Automated Detecting Arcus Senilis, Symptom for Presence Using Recognition Algorithm

Automated Detecting Arcus Senilis, Symptom for Cholesterol Presence Using Iris Recognition Algorithm

R.A.Ramlee, K.A.Aziz, S.Ranjit, Mazran Esro

Faculty of Electronic and Computer Engineering Universiti Teknikal Malaysia Melaka

Email: [email protected] Abstract centres for the entire body. Because of this anatomy and physiology, the eyes are Arcus senilis is a whitish ring-shaped or bow- in direct contact with the biochemical, shaped deposit in the . It is recognized as a sign of and is also associated hormonal, structural and metabolic to coronary heart disease (CHD). Iridology is processes of the body. This information is an alternative method to detect diseases using recorded in the various structures of the iris’s pattern observation. Iridologists believe eye, i.e. iris, , , cornea, that the whitish deposit on the iris is sign of and . Thus, it can be said that heart diseases. We develop the simple and the eyes are a reflex or window into the non-intrusive automation system to detect bioenergetics of the physical body and a cholesterol presence using iris recognition person’s feelings and thoughts [2]. There (image processing). This system applies iris recognition method to isolate the iris area, are a lot of arguments between iridologists normalization process and lastly determining (iridology’s practitioner) and the medical’s the cholesterol presence using OTSU histogram practitioner. Due to this argument, method to determine the threshold value. The numerous studies done by the medical’s result showed that the incidence of cholesterol practitioner found that the diagnosis done was high when eigen value exceeds a threshold by the iridologist upon the patient is not value. accurate. Another study on iris changes in related with disease, carried out in the Keywords: Iris recognition, biometric identification, iridology, Arcus Senilis, research of ocular complication of adult Corneal arcus, cardio diseases. rheumatoid arthritis done by [3], found that the mean duration of the arthritis and the mean duration of seropositivity I. INTRODUCTION were found to be significantly higher in patients with ocular (pigmented organ Iris is a pigmented, round, contractile in eye) complication [3]. Another study membrane of the eye, suspended between done on bilateral in the cornea and and perforated by acute myeloid leukaemia by [4], found the pupil depicted in Fig. 1. It regulates that ocular manifestations are common the amount of light entering the eye [1]. in patient with acute leukaemia. This According to [2], the eyes are connected can result from direction infiltration and continuous with the brain’s Dura by neoplastic cells of ocular tissues, mater through the fibrous sheath of the including , , retina, optic nerves, and they are connected iris and , or secondary to directly with the sympathetic nervous hematology abnormalities such as anemia, system and spinal cord. The optic tract thrombocytopenia, or hyperviscosity extends to the thalamus area of the brain. states or retinal destruction by This creates a close association with the opportunistic infection [4]. The history of hypothalamus, pituitary and pineal Iridology study on iris was done by the glands. These endocrine glands, within physician Philippus Meyens in 1670 in the brain, are major control and processing his book Chromatica Medica, explaining

ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 29

Automated Detecting Arcus Senilis, Symptom for Cholesterol Presence Using Iris Recognition Algorithm

R.A.Ramlee*, K.A.Aziz, S.Ranjit, Mazran Esro Universiti Teknikal Malaysia (UTeM), Faculty of Electronic Eng. and Computer Eng. Malaysia * email:[email protected]

Journal of Telecommunication, Electronic and Computer Engineering Abstract—Arcus senilis is a whitish ring-shaped or bow-shaped seropositivity were found to be significantly higher in patients deposit in the cornea. It is recognized as a sign of hyperlipidemia with ocular (pigmented organ in eye) complication [3]. and is also associated to coronary heart disease (CHD). Iridology Another study done on bilateral retinal detachment in acute is an alternative method to detect diseases using iris’s pattern myeloidthe leukaemia features by of [4], the found iris that(iridology). ocular manifestations In the the deposits consisting of cholesterol, observation. Iridologists believe that the whitish deposit on the book he stated that the eye (iris) contains cholesterol esters, phospholipids, and iris is sign of heart diseases. We develop the simple and non- are common in patient with acute leukaemia. This can result intrusive automation system to detect cholesterol presence using from directionvaluable infiltration information by neop lasticabout cells the of ocularbody. tissues, In triglycerides. The fatty acids that make iris recognition (image processing). This system applies iris including1881 optic a nerve, Hungrarian choroid, re physician,tina, iris and Dr.ciliary Ignatz body, or up many of the deposited lipid molecules recognition method to isolate the iris area, normalization process secondaryPeczley to hematology introduced abnormalities the first such chart as anemia,of the include palmitic, stearic, oleic, and and lastly determining the cholesterol presence using OTSU thrombocytopenia,iris explaining or hyperviscosity zone in the statesiris [2].or Theretinal linoleic acids [8]. histogram method to determine the threshold value. The result destruction by opportunistic infection [4]. The history of showed that the incidence of cholesterol was high when eigen idea begun when he found a dark scar in Iridology study on iris was done by the physician Philippus value exceeds a threshold value. the Owl’s iris scar turned white as the leg The current technique used to measure Meyens in 1670 in his book Chromatica Medica, explaining healed. the cholesterol level is by doing blood Index Terms— Iris recognition, biometric identification, the features of the iris (iridology). In the book he stated that iridology, Arcus Senilis, Corneal arcus, cardio diseases. the eye (iris) contains valuable information about the body. In test and the test is known as 1881 a TheHungrarian objective physician, of this Dr. project Ignatz Peczleyis to explain introduced profile. The lipoprotein profile is an I. INTRODUCTION the firsthow chart ofthe the irispresence explaining of zone cholesterol in the iris [2]. in The invasive method which causes discomfort Iris is a pigmented, round, contractile membrane of the eye, idea begunblood when vessel he found can a darkbe scardetected in the Owl’sby using iris scar amongst many patients. Reference [9] suspended between the cornea and lens and perforated by the turned whiteiris asrecognition the leg healed. algorithm. This method introduced a laser based technology as pupil depicted in Fig. 1. It regulates the amount of light The objectiveused the of [5],this project[6] iris is recognition to explain how methods the presence non-invasive technique to measure blood of cholesterol in blood vessel can be detected by using iris entering the eye [1]. According to [2], the eyes are connected and extends the study of eyes pattern to cholesterol through skin. They proposed recognition algorithm. This method used the [5], [6] iris and continuous with the brain’s Dura mater through the other application and in this case, the infrared (IR) absorption spectroscopic fibrous sheath of the optic nerves, and they are connected recognition methods and extends the study of eyes pattern to alternative medicine that is iridology as the characterization of cholesterol in directly with the sympathetic nervous system and spinal cord. other application and in this case, the alternative medicine that The optic tract extends to the thalamus area of the brain. This is iridology. the skin. Based on [10], skin contains creates a close association with the hypothalamus, pituitary approximately 11 percent by weight of and pineal glands. These endocrine glands, within the brain, all body cholesterol and when severe are major control and processing centres for the entire body. is present, the Because of this anatomy and physiology, the eyes are in direct numeric values obtained with the skin contact with the biochemical, hormonal, structural and cholesterol test increases. Thus, the palm metabolic processes of the body. This information is recorded test is not useful in identifying coronary in the various structures of the eye, i.e. iris, retina, sclera, artery disease and it is not intended to be cornea, pupil and conjunctiva. Thus, it can be said that the eyes are a reflex or window into the bioenergetics of the used as a screening tool to determine the physical body and a person’s feelings and thoughts [2]. There risk for coronary artery disease in general population. are a lot of arguments between iridologists (iridology’s Fig. 1: Anatomy practitioner) and the medical’s practitioner. Due to this argument, numerous studies done by the medical’s practitioner Fig. 1: Human Eye Anatomy In order to have a simple and non- IIris recognition is one of the most found that the diagnosis done by the iridologist upon the intrusive means to be as a screening tool Iris recognitionwidely implemented is one of the mostbiometric widely systemsimplemented patient is not accurate. Another study on iris changes in to detect cholesterol, we have considered biometricin systems use today. in use today.John JohnDaugman Daugman is is said to to have related with disease, carried out in the research of ocular alternative medicines. Iridology is one of complication of adult rheumatoid arthritis done by [3], found developedhave the mostdeveloped widely used the algorithms most widely and most used efficient the alternative medicines, which claims that the mean duration of the arthritis and the mean duration of methodsalgorithms of recognition, and but most there efficienthave been many methods new that iris pattern could reflect one’s health of recognition, but there have been and reveal the state of individual organs. many new findings and algorithms [7]. According to iridology, cholesterol in body Reference [5] has verified the uniqueness can be detected if there is a “sodium ring” of human iris patterns and developed an in the patient’s eyes. However, since there open source iris recognition system. were statements that regard iridology as medical fraudulent [6] we were looking Based on the iris recognition methods at other medical statements that can and iridology chart, a MATLAB program relate cholesterol and other organs. We has been created to detect the present of found out that high cholesterol can be cholesterol in our body. detected from changes in iris pattern and they are called Arcus Lipoides (Arcus Cholesterol or is Senilis or Arcus Juvenilis). Arcus senilis is a high level of lipid in the blood poses a greyish or whitish arc or circle visible a significant threat to person’s health. It around the peripheral part of the corner is an indication of elevated cholesterol in older adults. Arcus senilis is caused which may lead to cardiovascular by lipid deposits in the deep layer of the diseases. It is caused by extracellular lipid peripheral cornea and not necessarily deposition in the peripheral cornea, with

30 ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 Automated Detecting Arcus Senilis, Symptom for Cholesterol Presence Using Iris Recognition Algorithm associated with high blood cholesterol. is nervous or sensory. The fluids in the However, similar discoloration in the eye are divided by the lens into the eyes of younger adults (arcus juvenilis) vitreous humor (behind the lens) and is often associated with high blood the aqueous humor (in front of the lens). cholesterol [11]. This statement proves The lens itself is flexible and suspended that iris pattern can be analyzed and used by ligaments which allow it to change as another technique to detect cholesterol shape to focus light on the retina, which presence in body. is composed of sensory neurons (NLM, 2010). Fig. 1 shows the anatomy of human Reference [12], conclude in his study eye which contain the area of sclera and the presence of Arcus Senilis before iris for references. The iris image needs the age of 56 and large wrist size were to be extracted from the original eye found to appear with a frequency in image. This solid iris image is used in coronary group which made their this system to verify the presence of presence statistically significant at cholesterol. Thus it is vital to isolate this level 5%. Hypercholesterolemia was part (iris) from the whole unwanted part common finding in coronary patient who in the eye (sample). This separation or demonstrated Arcus Senilis and greying of segmentation is the process of remove hair. According to [11] although iridology the outer part of the eye (outside the iris has been criticized as an unfounded circle), in order to get solid image of iris diagnostic tool, many iridologists are that useful for localisation the cholesterol presently practicing in many areas. In lipid. Generally this eye breaks up into Germany, 80% of Heilpraktiker (non- two parts, the first part is the inner region medically qualified health practitioners) which is the iris and pupil boundary and practice iridology [13]. In this study, the second part is the outer regions, the [11] investigated the ACE genotypes of iris and sclera boundary. The quality of hypertensive patients classified by their the images is very important to get the iris constitutions. As a result, 74.7% of best result, thus the images should not hypertensive patients were neurogenic have any impurities that can cause miss or cardio-renal connective tissue localization. These impurities include the weakness type. Also, the frequencies of flash reflection from camera and wrong DD genotype were significantly higher angle of image capture. in hypertensive patients than in controls. These results are consistent with the In this project the sample of eye is very reports that DD genotype was associated important because analysis base on the with hypertension. Therefore, [11] data from human eyes. For this work, present the results support that D allele we focus on the detecting arcus senilis is a candidate gene for hypertension, and using iris recognition algorithm and suggest an apparent relationship between not focusing of method to extract the ACE genotype and iris constitutions, as iris images. Therefore, we use images well as the novel possibility of molecular obtained freely from a few free database genetics understanding of iridology. sources that can found in website, these databases such as UBIRIS, UPOL, MMU and CASIA [14-17]. II. EYE IMAGE UBIRIS [14] database is comprised of The eye is the organ of sight, a nearly 1877 images collected from 241 subjects spherical hollow filled with fluids within the University of Beira Interior 6 (humors). The outer layer or tunic in two distinct sessions and constituted, (sclera, or white, and cornea) is fibrous at its release date, the world’s largest and protective. The middle tunic layer public and free iris database for biometric (choroid, ciliary body and the iris) is purposes. vascular. The innermost layer (the retina)

ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 31 date, the world’s largest public and free iris database for simple geometric objects, such as lines and circles, present in biometric purposes. an image. The circular Hough transform can be employed to In CASIA [17], iris image database includes 756 iris images deduce the radius and centre coordinates of the pupil and iris from 108 eyes, hence 108 classes. For each eye, 7 images are regions. An automatic segmentation algorithm based on the captured in two sessions, where three samples are collected in circular Hough transform is employed [18]. Firstly, an edge the first and four in the second session. Similarly to the above map is generated by calculating the first derivatives of described database, its images were captured within a highly intensity values in an eye image and then set the threshold constrained capturing environment, which conditioned the characteristics of the resultant images. They present very close base on the result. From the edge map, votes are cast in Hough and homogeneous characteristics and their noise factors are space for the parameters of circles passing through each edge exclusively related with iris obstructions by and point. These parameters are the centre coordinates xc and yc, (Fig. 2). Moreover, the post process of the images and the radius r, which are able to define any circle according filled the pupil regions with black pixels, which some authors to the equation used to facilitate the segmentation task. x 2 + y 2 + r 2 = 0 (1) c c

A maximum point in the Hough space will correspond to the radius and centre coordinates of the circle best defined by Fig. 2: Examples of iris images from the CASIA [17] database. the edge points. References [18], [19] also make use of the parabolic Hough transform to detect the eyelids, The Multimedia University has developed a small data set approximating the upper and lower eyelids with parabolic of 450 iris images MMU [16]. These images were captured arcs, which are represented as; using LG IrisAccessR 2200 camera. This is a semi-automated 2 camera that operates at the range of 7-25 cm. Obviously, the ((−x−hj )sinθj +(y−kj )cosθj ) =aj ((x−hj )cosθj +(y−kj )sinθj ) (2) images are highly homogeneous and their noise factors are exclusively related with small iris obstructions by eyelids and eyelashes (Fig. 3). a (h ,k ) j controls the curvature, j j is the peak of the parabola θ and j is the angle of rotation relative to the x-axis.

In performing the preceding edge detection step, [18] bias

the derivatives in the horizontal direction for detecting the Fig. 3: Examples of iris images from the MMU [16] database. eyelids, and in the vertical direction for detecting the outer circular boundary of the iris. The motivation for this is that the Journal of Telecommunication, Electronic and Computer EngineeringFor this project the difficulty is to achieve the subject or eyelids are usually horizontally aligned, and also the patient to get real image of eye sample, the best place to get edge map will corrupt the circular iris boundary edge map if these real eye images is at department since using all gradient data. Taking only the vertical gradients for this department deal with various case of eye problem that locating the iris boundary will reduce influence of the eyelids later can be refer to Arcus Senilis problem. For this reason, the In CASIA [17], iris image database sampleswebsite only cansuch be used as from National free source medicalLibrary website. of when performing circular Hough transform, and not all of the includes 756 iris images from 108 eyes, Fig.Medicine 4 shows a andfew samplesMediscan from medical clipart website library. such as edge pixels defining the circle are required for successful National Library of Medicine and Mediscan clipart library. localisation. Not only does this make circle localisation more hence 108 classes. For each eye, 7 images accurate, it also makes it more efficient, since there are less are captured in two sessions, where edge points to cast votes in the Hough space. three samples are collected in the first IV. SEGMENTATION and four in the second session. Similarly This segmentation (localization) process is to search for the to the above described database, its centre coordinates of the pupil and the iris along with their radius. These coordinates are marked as ci, cp, where ci images were captured within a highly represented as the parameters of [xc,yc,r] of the limbic and iris constrained capturing environment, Fig. 4: Examples4: Examples of iris images of from iris medical images website. from medical date, the world’s largest public and free iris database for simple geometric objects, such as lines and circles, presentboundary in and cp represented as the parameters of [xc,yc,r] of the pupil boundary. It makes use of [18] to select the possible biometricwhich purposes. conditioned the characteristics of an image. The circularIII. HOUGHwebsite. Hough TRANSFORM transform can be employed to centre coordinates first. The method consist of threshold The Hough transform is a standard computer vision Inthe CASIA resultant [17], iris imageimages. database They includes present 756 iris very images deduce the radius and centre coordinates of the pupil andfollowed iris by checking if the selected points (by threshold) fromclose 108 eyes, and hence homogeneous 108 classes. For eachcharacteristics eye, 7 images are algorithmregions. that An canautomatic be used segmentationto determine thealgorithm parameters based of oncorrespond the to a local minimum in their immediate capturedand in their two sessions, noise where factors three samplesare exclusively are collected in circular Hough transform is employed [18]. Firstly, an edge the first and four in the second session. Similarly to the above III. hOUGH TRANSFORM map is generated by calculating the first derivatives of describedrelated database, with its iris images obstructions were captured bywithin eyelids a highly intensity values in an eye image and then set the threshold constrainedand eyelashes capturing (Fig.environment, 2). Moreover, which conditioned the post the The Hough transform is a standard characteristicsprocess of of the theresultant images images. filledThey present the very closepupil computerbase on the result. vision From thealgorithm edge map, votes that are castcan in Hough space for the parameters of circles passing through each edge anddate, regionshomogeneous the world’s with characteristicslargest black public pixels, andand freetheir whichiris noise database factors some for are simple be geometricused objects,to determine such as lines andthe circles, parameters present in point. These parameters are the centre coordinates x and y , exclusivelybiometric purposes.related with iris obstructions by eyelids andan ofimage. simple The circular geometric Hough transform objects, can be employed such to asc c authors used to facilitate the segmentation and the radius r, which are able to define any circle according eyelashesIn CASIA (Fig. [17], 2). irisMoreover, image database the post includes process 756 of iris the images images deduce lines the radiusand andcircles, centre coor presentdinates of thein pupilan andimage. iris filledfromtask. the 108 pupil eyes, regions hence 108with classes. black Forpixels, each which eye, 7 someimages authors are regions. to the An equation automatic segmentation algorithm based on the usedcaptured to facilitate in two thesessions, segmentation where three task. samples are collected in circularThe Hough circular transform Hough is employed transform [18]. Firstly, ancan edge be the first and four in the second session. Similarly to the above 2 2 2 mapemployed is x generated+ y + bytor calculating=deduce0 the the first radiusderivatives andof (1) described database, its images were captured within a highly c c intensity values in an eye image and then set the threshold constrained capturing environment, which conditioned the centre coordinates of the pupil and iris characteristics of the resultant images. They present very close baseregions. onA the maximum result. AnFrom point theautomatic edgein the map, Hough votes aresegmentationspace cast willin Hough correspond to space for the parameters of circles passing through each edge and homogeneous characteristics and their noise factors are algorithmthe radius and based centre coordinateson the circularof the circle Hough best defined by exclusively related with iris obstructions by eyelids and point. These parameters are the centre coordinates xc and yc, Fig. 2: ExamplesFig. 2: ofExamples iris images from of iristhe CASIA images [17] database.from the the edge points. References [18], [19] also make use of the date, thedate, world’s the largestworld’seyelashes publiclargest (Fig. 2).and publicMoreover, free andiris th e freedatabasepost processiris databasefor of the simpleimages for geometricandsimpletransform the radius geometric objects, r, which is such objects, areemployed ableas lines tosuch define and as [18].any linescircles, circle andFirstly, according present circles, an in present in CASIA [17] database. parabolic Hough transform to detect the eyelids, biometricbiometric purposes. purposes. filled the pupil regions with black pixels, which some authorsan image. toan Thetheedge image. equation circular map The Hough circularis generated transform Hough transform canby be calculating employed can be employed to to The Multimedia University has developed a small data set approximating the upper and lower eyelids with parabolic In CASIAIn [17], CASIA irisused [17],image to facilitateiris database image the segmentationdatabaseincludes 756includes task. iris images 756 iris imagesdeduce thededuce theradius first the and radius centrederivatives and coor centredinates of coor intensity ofdinates the pupil ofvalues theand pupil iris in and iris of 450 iris images MMU [16]. These images were captured arcs,2 which2 2 are represented as; (1) from 108from eyes, 108 hence eyes, The108 hence classes. Multimedia 108 For classes. each University Foreye, each 7 images eye, has 7are developedimages are xc + yc + r = 0 using LG IrisAccessR 2200 camera. This is a semi-automatedregions. An regions.an automatic eye An image automatic segmentation and segmentation then algorithm set the algorithm based threshold on based the on the captured in two sessions,a small where datathree samplesset of 450are collectediris images in MMU captured in two sessions, where three samples are collectedcircular in Hough transform is employed2 [18]. Firstly, an edge camera that operates at the range of 7-25 cm. Obviously, the circularAbase maximum((− xHoughon−hj )sinpointtheθ jtransform +inresult.(y −thekj )Houghcos θFromisj ) employed=spaceaj (( x thewill−hj ) coscorrespond edge[18].θj +( y Firstly,−map,k toj )sin θ j )an (2)edge the first andthe fourfirst andin the four[16]. second in theThese session. second images Similarlysession. were Similarly to the captured above to the above using images are highly homogeneous and their noise factorsmap isare the mapgeneratedvotes radius is and generated are bycentre castcalculating coordinates byin calculatingHough ofthe the circlefirst space thebestderivatives definedfirst for derivativesbythe of of describeddescribed database, database, itsLG images IrisAccessR its were images captured were 2200 capturedwithin camera. a withinhighly This a highly is a exclusivelyFig. 2: Examples related of withiris images small from iris the CASIAobstructions [17] database. by eyelidsintensity and the intensityvalues edge points.in values an Referenceseye in image an eye[18] and , image[19] then also and setmake thenthe use threshold setof the the threshold constrainedconstrained date,capturing the world’scapturing environment, largest environment, public which and conditioned whichfree iris conditioned database the for thesimple parameters geometric objects, of suchcircles as lines passing and circles, throughpresent in eyelashessemi-automated (Fig. 3). camera that operates parabolic Hough transform to detect the eyelids, biometric purposes. base on thebase result. ona j the From result. the Fromedge map,the edge (voteshj ,k jmap,) are votescast in are Hough cast in Hough characteristicscharacteristics of the resultant Theof theMultimedia resultantimages. University They images. present has They developed very present close a small very data close set an image.approximatingeach The controlsedge circular the upperthepoint. Hough curvature, and transformTheselower eyelids parameterscan is thewithbe peakemployed parabolic of theare to parabola In CASIAat [17], the iris range image databaseof 7-25 includes cm. 756Obviously, iris imagesspace the deducefor space the the parameters forradius the andparameters of centre circles coor of passingdinates circles ofthrough passing the pupil eachthrough and edge iris each edge and homogeneousand homogeneous characteristicsof 450 iris characteristics images and MMU their [16].and noise Thesetheir factors imagesnoise are werefactors captured are arcs,the which θcentrej are represented coordinates as; xc and yc, and from 108 eyes,date,images hencethe world’s 108are classes. largesthighly Forpublic eachhomogeneous and eye, free 7 imagesiris database are and for simpleand geometric is the angleobjects, of such rotation as lines relative and circles, to the presentx-axis. in exclusivelyexclusively related usingwithrelatedbiometric LGiris IrisAccessR withpurposes. obstructions iris 2200obstructions cameraby eyelids. This byis a andsemi-automatedeyelids point. andregions. Thesepoint.anthe image. parametersAn Theseradius automaticThe circularparameters r,are whichsegmentation Houghthe centre are transformare the ablealgorithmcoordinates centre can to be coordinatesdefineemployedbased xc andon to theany y cx, c and yc, captured incamera two sessions, that operates where at thethree range samples of 7-25 are cm. collected Obviously, in the − − θ + − θ 2 = − θ + − θ (2) theirIn CASIA noise [17], factors iris image aredatabase exclusively includes 756 iris relatedand images thecircular radiusdeduce(( xHough r, hthej )whichsin radius transformj (y areandkj ) cosablecentre isj ) employedto coora jdefine((xdinateshj )cos any[18]. ofj the ( circleyFirstly, pupilkj )sin according and j an) irisedge eyelasheseyelashes (Fig.the first2). (Fig.andMoreover, four 2). in Moreover, thethe second post processthsession.e post Similarly ofprocess the images toof the the above images andcircle theIn radiusperforming according r, which the topreceding are the able equation edgeto define detection any step,circle [18] according bias filled the pupil regionsimagesfromwith with 108 are blacksmall eyes,highly pixels,hence homogeneousiris 108 whichobstructions classes. someand For their each authors noiseeye, by 7 factors imageseyelids are aremap regions. is generated An automatic by calculatingsegmentation the algorithm first derivativesbased on the of filleddescribed the pupil database,captured regions inits two withimages sessions, black were where pixels, captured three which samples within some area highlycollected authorsto the in equation to thethe equation derivatives in the horizontal direction for detecting the exclusivelyand eyelashes related with (Fig. small iris3). obstructions by eyelids andintensity circular values Hough in transform an eye imageis employed and then[18]. setFirstly, the anthreshold edge used to facilitateusedconstrained to facilitatethe segmentationFig.eyelashes thecapturing 3: first Examplesthe and(Fig.segmentation four environment,of3). task. iris in imagesthe second task.from which thesession. MMU conditioned Similarly [16] database. to the the above eyelids, and in the vertical direction for detecting the outer mapa j is generated by calculating(hj ,k j ) the first derivatives of characteristicsdescribed of the database,resultant itsimages. images They were presentcaptured very within close a highly 2base on2circular the controls2 result.2 boundary 2the From curvature,2 the of edgethe iris. map, isThe the votes motivationpeak are of castthe parabolafor in Houghthis(1) is that the(1) xc + yintensityc x+c r+ valuesy=c 0 +inr an =eye0 image and then set the threshold Forconstrained this project capturing the difficulty environment, is to which achieve conditioned the subject thespace or forθ j the parameters of circles passing through each edge and homogeneous characteristics and their noise factors are andbaseeyelids on is the result.angleare usually ofFrom rotation the horizontally edge relative map, to votes the aligned, x-axis. are cast and in Houghalso the eyelid patientcharacteristics to get real of image the resultant of eye images. sample, They the present best placevery close topoint. get These parameters are the centre coordinates x and y , exclusively related with iris obstructions by eyelids and spaceedge for mapthe parameters will corrupt of circles the circular passing iristhrough boundary eachc edge edge c map if eyelashesthese (Fig.and real homogeneous2). eye Moreover, images characteristics isth eat post ophthalmology process and their of the noisedepartment images factors A are sincemaximumand theInA radiusperforming maximum point r, which inthe thepoint preceding are Hough able in thetoedge spacedefine Hough detection anywill space circle step,correspond [18] willaccording bias correspond to to point.using These all parametersgradient data.are the Taking centre onlycoordinates the vertical xc and gradientsyc, for this exclusivelydepartment related deal withwith variousiris obstructions case of eyeby eyelidsproblemthe andradius that the the andA derivativesradius maximumcentre and coordinatesin centrethe horizontalpoint coordinates of inthedirection thecircle of forHoughthe best detecting circle defined space bestthe by defined by filled the pupileyelashes regions (Fig. with 2). blackMoreover, pixels, the whichpost process some ofauthors the images to theand equationlocating the radius the r, whichiris boundary are able to will define reduce any circleinfluence according of the eyelids used to laterfacilitate Fig.canFig. 3: bethe3: Examples referExamples segmentation toof irisArcus images of Senilistask. fromiris the images problem. MMU [16] from database. For this the reason, MMU the eyelids, will and correspond in the vertical todirection the radiusfor detecting and the centreouter Fig. 2: ExamplesFig. 2: of Examples iris imagesfilled of the irisfrom pupil images the regionsCASIA from thewith[17] CASIA database.black pixels,[17] database. which somethe authors edge thetopoints. the whenedge equation Referencesperforming points. References circular[18], [19] Hough [18] also ,transform, [19] make also use and make ofnot the alluse of ofthe the samplesused toonly facilitate can bethe usedsegmentation[16] from database. free task. source medical website.circular 2 boundary2 2 of the iris. The motivation for this is that the(1) For this project the difficulty is to achieve the subjectparabolic or xparabolic c coordinates+Houghyc + r Houghtransform= 0 of transform theto circledetect to bestthedetect defined eyelids,the eyelids, Fig. 4 shows a few samples from medical website such aseyelids edge2 are pixels 2usually2 defining horizontally the aligned,circle areand requiredalso the eyelidfor successful The MultimediaThe Multimedia University University has developed has developed a small dataa small set data set x + y + r = 0 (1) Nationalpatient Libraryto get real of Medicineimage of eye an dsample, Mediscan the bestclipart place library.approximating to get edgeapproximatingbylocalisation. cmap thethe willc upper corruptedge Notthe and the only upper points.circular lowerdoes and this iriseyelids boundarylowerReferencesmake circlewitheyelids edge localisationparabolic map with[18], if parabolic more theseFor real this eye project images is theat ophthalmology difficulty department is to achieve since of 450 irisof images450 iris MMU images [16]. MMU These [16]. images These were images captured were capturedarcs, which Ausingarcs, maximum[19] areaccurate, allwhich represented gradientalso pointareit alsodata. represented makein as;makes theTaking Hough use itonly moreas; spacethe of efficient,vertical thewill gradientscorrespond sinceparabolic there for to are less this department deal with various case of eye problem that A maximum point in the Hough space will correspond to using LGusing IrisAccessR LG IrisAccessR the2200 subject camera 2200. Thisorcamera patient is a. Thissemi-automated tois aget semi-automated real imagethe locatingradius and the iriscentre boundary coordinates will reduce of the influence circle bestof the defined eyelids by later can be refer to Arcus Senilis problem. For this reason, the theHough edgeradius points and transformcentre to cast coordinates votes into ofthe detectthe Hough circle space.bestthe defined eyelids, by Fig. 2: Examplesof eye of iris sample, images from the CASIA best [17] place database. to get these−the− whenedgeθ performingpoints.+ − References circularθ 2 = Hough [18]− ,transform, 2[19] alsoθ + and make− not alluse θof of the the (2) (2) camera thatcamera operates that samplesoperatesat Fig.the 2: raonly Examples ngeat canthe of of be irisra7-25 usednge images cm.fromof from 7-25 Obviously, thefree CASIA sourcecm. [17] Obviously, medical database.the website. the(( x hjthe)sinapproximating ((edge−jx− (hpoints.yj )sinkj )θ cosjReferences+(yj )−k ja)thecosj (([18]xθ j,)h [19]jupper=)cosa j ((alsoxj − (makehyj )andcosk juse)θsinj + of(lowerjy )the−k j )sin θj ) parabolicedge pixels Hough defining transform the circle areto requireddetect forthe successful eyelids, images areimages highly are homogeneous Fig.highlyreal 4 shows homogeneouseye a andfewimages samplestheir and noise isfrom theirat medicalfactors noiseophthalmology websiteare factors such are as paraboliceyelids Hough with transform parabolicIV. StoEGMENTATION detectarcs, thewhich eyelids, are The MultimediaNational Library University of Medicine has developed and Mediscan a small clipart data library. set localisation. Not only does this make circle localisation more departmentThe Multimedia Universitysince this has developeddepartment a small data deal setapproximating approximatingThis segmentationthe the upper upper and and(localization) lowerlower eyelids process withwith isparabolic parabolicto search for the exclusivelyexclusively ofrelated 450 iriswith related images small with MMU iris small obstructions [16]. iris These obstructions byimages eyelids were by and eyelidscaptured and accurate,represented it also makes as; it more efficient, since there are less of 450 iris images MMU [16]. These images were capturedarcs, arcs, whichcentre which are coordinatesare represented represented ofas; as; the pupil and the iris along with their eyelasheseyelashes (Fig. 3). (Fig.with 3). various case of eye problem that edge points to cast votes in the Hough space. using LG IrisAccessRusing LG IrisAccessR 2200 camera 2200. cameraThis is. Thisa semi-automated is a semi-automated a j a j (hj ,k j ) (hj ,k j ) controlsradius. thecontrols curvature,These the coordinates curvature, is arethe peakmarked is ofthe asthe peak cparabolai, ofcp, thewhere parabola ci camera thatcameralater operates canthat at operates bethe referra ngeat the ofto ra 7-25 Arcusnge ofcm. 7-25 Senilis Obviously, cm. Obviously, problem. the the − −((−x−h )θsin+θ +−(y−k )cosθθ2)2==a ((x−−h )cosθθ ++(y−−k )sinθ θ) (2)(2) (( x representedhj )sinj j (j y askj )thecosj parametersj )j aj ((j x h ofjj)cos [x j,yj ,r(y] ofjkj )thesinj limbicj ) and iris Fig. 4: Examples of iris images from medical website. θ j θ IV. SEGMENTATION c c images are highly homogeneous and their noise factorsand are is theand anglej is theof rotationangle of relative rotation to relative the x-axis. to the x-axis. images are Forhighly this homogeneous reason, theand theirsamples noise factorsonly canare be boundary and cp represented as the parameters of [xc,yc,r] of exclusively exclusivelyrelated with related small with iris small obstructions iris obstructions by eyelids by eyelids and and This segmentation (localization) process is to search for the used from III.freeH OUGHsource TRANSFORM medical website. centrethe coordinates pupil boundary. of the pupilIt makes and usethe irisof [18]along to with select their the possible eyelashes (Fig. 3). a (h ,k ) eyelashes (Fig. 3). In performinga Incentrej controlsperforming the coordinates thepreceding curvature, the precedingfirst. ( edgeh ,j k Thej) detectionis the methodedge peak detectionofstep, consistthe parabola[18] step,of bias threshold [18] bias TheFig. Hough 4 shows transform a few is samplesa standard from computer medical vision radius. j controls These the coordinates curvature, arej markedj is the as peak ci, ofcp , thewhere parabola ci the derivativesfollowedθ in the by horizontalchecking ifdirection the selected for pointsdetecting (by thethreshold) Fig. 4: Examples of iris images from medical website. representedtheand derivativesj is the as angle the parameters ofin rotation the horizontalrelativeof [xc,y cto,r] the of directionx-axis.the limbic andfor iris detecting the algorithm that can be used to determine the parameters ofθ j Fig. 3: ExamplesFig. 3: of Examples iris images of irisfrom images the MMU from [16] the MMUdatabase. [16] database. eyelids,and andboundaryeyelids, iscorrespond inthe the angleandand verticalcp ofin represented to rotationthe a direction verticallocal relative as the midirection parametersfortonimum the detecting x-axis. forofin [ x detectingctheir,ythec,r ] outerofimmediate the outer III. HOUGH TRANSFORM circular boundarythe Inpupil performing boundary. of the the iris. It precedingmakes The use motivation edge of [18] detection to selectfor step, this the [18]ispossible that bias the centrecircular coordinates boundary first. of theThe iris.method The consistmotivation of threshold for this is that the For this Forproject this the project Thedifficulty Houghthe difficulty istransform to achieve isis toa theachievestandard subject thecomputer orsubject vision or Inthe performing derivatives the in precedingthe horizontal edge direction detection for step,detecting [18] the bias 32 ISSN: 2180 - 1843 eyelids Vol. 3 are followed eyelids No. usually 2 by are July-December checking horizontallyusually if horizontally the aligned,selected 2011 pointsandaligned, also(by threshold)andthe alsoeyelid the eyelid patient to get real imagealgorithmFig. of 3: eyeExamples that sample,can of irisbe images used the from tobest determinethe MMUplace [16] tothe database. get parameters ofthe derivativeseyelids, and in thethe verticalhorizontal direction direction for detecting for detecting the outer the patient to get real image of eye sample, the best place toedge get mapcorrespondedge will mapcorrupt towill athe corruptlocal circular mithenimum iriscircular boundaryin iristheir boundary edgeimmediate map edge if map if these realthese eye Fig. realimages 3: Exampleseye isimages at of irisophthalmology images is at from ophthalmology the MMU department [16] database. department since sinceeyelids, circular and boundary in the ofvertical the iris. direction The motivation for detecting for this is thethat theouter For this project the difficulty is to achieve the subjectusing or all usingeyelidsgradient allare data.gradientusually Taking horizontally data. only Taking aligned, the onlyvertical and thealso gradientsvertical the eyelid gradients for for this department deal with various case of eye problem that circular boundary of the iris. The motivation for this is that the this departmentFor this patientproject deal to the getwith difficultyreal various image isof tocaseeye achieve sample, of eye thethe problembestsubject place or thatto get edge map will corrupt the circular iris boundary edge map if these real eye images is at ophthalmology departmentlocating sinceeyelids thelocating irisare usuallyboundary the iris horizontally boundary will reduce aligned,will influence reduce and influence alsoof the the eyelidseyelid of the eyelids later can laterbe patientrefer can to beto Arcus referget real Senilisto imageArcus problem. ofSenilis eye sample, problem.For this the reason, Forbest this place the reason, to get the using all gradient data. Taking only the vertical gradients for this department deal with various case of eye problemwhen that edgeperforming when map willperforming circular corrupt Houghthecircular circular transform, Hough iris boundary transform, and not edge alland mapof not the if all of the samples onlysamplesthese can real onlybe eyeused can images frombe used isfree at from sourceophthalmology free medical source department website.medical website.since locating the iris boundary will reduce influence of the eyelids later can be refer to Arcus Senilis problem. For this reason,edge the usingpixels edge all defining gradientpixels data.definingthe circleTaking the areonly circle required the verticalare requiredfor gradients successful for for successful Fig. 4 showsFig.this 4a departmentshowsfew samplessamples a few deal only samplesfromwith can medicalvariousbe usedfrom fromcase websitemedical freeof eyesource suchwebsite problem medical as such that website. as when performing circular Hough transform, and not all of the locating the iris boundary will reduce influence of the eyelids National NationalLibrarylater can of Library Medicinebe Fig.refer of4 toshows Medicine anArcusd Mediscana Senilisfew an samplesd problem. Mediscan clipart from For library.medical clipart this reason, website library. thelocalisation.such as localisation.edge Not pixels only defining Notdoes only thethis circle doesmake arethis circle required make localisation circlefor successful localisation more more samples onlyNational can be Library used offrom Medicine free sourceand Mediscan medical clipart website. library.accurate, when accurate,localisation.it performing also makes it Not alsocircular only it makesmore does Hough this efficient,it makemoretransform, circle efficient,since localisationand there not since all are more of there lessthe are less edgeaccurate, pixels itdefining also makes the itcircle more efficient,are required since therefor successfulare less Fig. 4 shows a few samples from medical website such edgeas pointsedge to pointscast vo totes cast in the vo tesHough in the space. Hough space. National Library of Medicine and Mediscan clipart library. localisation.edge points Not to castonly vo doestes in this the Houghmake space.circle localisation more accurate, it also makes it more efficient, since there are less IV. SIV.EGMENTATIONSIV.EGMENTATIONSEGMENTATION edge points to cast votes in the Hough space. This segmentationThis segmentation segmentation (localization) (localization) (localization) process process is isprocess toto search is for tofor thesearch the for the centre coordinates of the pupil and the iris along with their centre coordinatescentre coordinates of theIV. pupil ofS EGMENTATIONthe and pupil the irisand alongthe iris with along their with their radius. These coordinates are marked as ci, cp, where ci

radius. ThisTheseradius.represented segmentation coordinates These as the parameterscoordinates(localization) are marked of [ x cprocessare,yc,r ] asmarkedof theisc ito, limbic csearch pas, whereandc ifor, irisc the p,c i where ci Fig. 4: Examples of iris images from medical website. Fig. 4: ExamplesFig. 4: of Examples iris images of irisfrom images medical from website. medical website. representedcentrerepresentedboundary coordinatesas the and parameters cpas of representedthe the parameters pupilof [asx ca ,ythendc,r parametersofthe] of [xiris thec,y calong ,rlimbic of] of[x cwiththe,y cand,r ] limbic theirof iris and iris the pupil boundary. It makes use of [18] to select the possible III. HOUGH TRANSFORM boundaryradius.boundary and These cp represented andcoordinates cp represented asare the marked parameters as theas parametersci , ofcp ,[ xwherec,yc,r of] ofc[i x c,yc,r] of centre coordinates first. The method consist of threshold The Hough transform is a standard computerthe vision pupilrepresented boundary. as the It parametersmakes use of of [ x[18]c,yc,r ]to of select the limbic the possible and iris Fig. 4: ExamplesIII. H OUGHof irisIII. images THRANSFORMOUGH from medical TRANSFORM website. thefollowed pupil by boundary. checking ifIt themakes selected use points of [18] (by tothreshold) select the possible algorithm that can be used to determine the parameterscentre ofboundary coordinatescentre and coordinates cp first. represented The first. methodas theThe parameters consistmethod ofconsist [xthresholdc,yc,r ]of of threshold The HoughThe transformHough transform is a standard is a standardcomputer computer vision vision correspond to a local minimum in their immediate III. HOUGH TRANSFORM followedthe pupilfollowedby checkingboundary. by checking Itif makesthe selected useif ofthe [18] selectedpoints to select (by points the threshold) possible (by threshold) algorithmalgorithm that can thatbe usedcan beto determineused to determine the parameters the parameters of ofcentre coordinates first. The method consist of threshold The Hough transform is a standard computer visioncorrespond correspond to a localto a milocalnimum mi nimumin their in immediatetheir immediate followed by checking if the selected points (by threshold) algorithm that can be used to determine the parameters of correspond to a local minimum in their immediate Automated Detecting Arcus Senilis, Symptom for Cholesterol Presence Using Iris Recognition Algorithm

neighbourhood these points serve as the possible centre coordinates for the iris. These radius values were set manually, was set manually, depending on the a j (h j ,k j ) (Table 1) shows the rmin and rmax varies from 75 to 250 controls the curvature, is thewhere database the ranges used. for iris For between the CASIA90 to 150 database, pixels and for peak of the parabola and is the angle ofpupil rminradius isare set 28 toto 7555 pixels.pixels The and input rmax for isthis set function to is rotation relative to the x-axis. the image160 pixels.to be segmented The sample and the of input Arcus parameters Senilis in this functioneye including(arcus7.bmp) rmin andis downloadingrmax (the minimum from and In performing the preceding edgemaximum (Mediscan, values of 2000),the iris forradius).The this image range theof radius rmin values detection step, [18] bias the derivativesto searchis set for to was 70 setpixels manually, and rmixdepending is set on tothe 260 database Fig. 5: An example where segmentation fails. The segmentation is failing in the horizontal direction for detectingused. For the CASIA database, rmin is set to 55 pixels and to detection correctly the edges of the pupil border, but it segmented pixels. Table 1 shows the others sample illumination light. the eyelids, and in the vertical directionrmax is set to 160 pixels. The sample of Arcus Senilis eye (arcus7.bmp)run for is segmentation downloading from process. (Mediscan, 2000), for this for detecting the outer circular boundary image the rmin is set to 70 pixels and rmix is set to 260 pixels. But luckily the significant area of white ring (Arcus Senilis) of the iris. The motivation for this isTable 1 showsTable the others1: shows sample the runexample for segmentation of the process. lay at the boundary of sclera or iris up to the pupil, so as long that the eyelids are usually horizontally images and their rmin and rmax values. as the segmentation done correctly on the iris it can considered aligned, and also the eyelid edge map will succeed. This segmentation image will be crop base on the Image rmin rmax corrupt the circular iris boundary edge value of iris’s radius. Iris1.bmp (CASIA) 75 260 map if using all gradient data. Taking Iris2.bmp (CASIA) 75 260 only the vertical gradients for locating Iris3.bmp (CASIA) 75 260 the iris boundary will reduce influence Iris4.bmp (CASIA) 75 260 Iris6.bmp (CASIA) 75 260 of the eyelids when performing circular c_mo1.bmp 50 83 Hough transform, and not all of the edge c_mo2.bmp 75 260 pixels defining the circle are required for normal1.bmp 75 260 Arcus1.bmp (Medical Web) 80 260 successful localisation. Not only does this Arcus2.bmp (Medical Web) 80 260 make circle localisation more accurate, it Arcus3.bmp (Medical Web) 80 260 also makes it more efficient, since there Arcus4.bmp (Medical Web) 80 260 Arcus5.bmp (Medical Web) 80 260 Fig. 6: Another example where segmentation fails to find the edges of the are less edge points to cast votes in the Arcus6.bmp (Medical Web) 80 260 pupil border. Hough space. Arcus7.bmp (Medical Web) 80 260 Arcus2.bmp (Medical Web) 80 260 The problem with this image is the existing of illumination Arcus2.bmp (Medical Web) 80 260 from camera (on top of the pupil) will cause inaccurate Arcus2.bmp (Medical Web) 80 260 detection of iris and pupil boundary. This process consider IV. SEGMENTATION ubiris1.bmp (UBIRIS) 85 260 ubiris2.bmp (UBIRIS) 85 260 failure if the purpose of this segmentation process is used for This segmentation (localization) process is ubiris2.bmp (UBIRIS) 85 260 biometric iris recognition, but for this system since the image to search for the centre coordinates of the only need to be analyze in 30 percent from the limbic (border Table 1: shows the example of the images and their rmin and rmax values. of iris) so this result can be used. pupil and the iris along with their radius. The output of this function will be the These coordinates are marked as ci, cp, The output of this function will be the value of cp and ci, value of cp and ci, which is the value of V. NORMALIZATION where ci represented as the parameters ofwhich is the value of [xc, yc, r] for the pupilary boundary and [x , y , r] for the pupilary boundary and After the iris is localized, the next step is normalization (iris [xc,yc,r] of the limbic and iris boundarythe limbicc / ciris boundary and also the segmented image. The the limbic / iris boundary and also the enrolment). It is a process after localization (segmentation) and cp represented as the parameters ofprogram processes the image from [17] database as shown in the Fig.segmented 5. This image.image gives The programwrong detection processes on pupil intends to change the iris region to the fixed form (circle [xc,yc,r] of the pupil boundary. It makes boundarythe becauseimage segmentationfrom [17] databaseon pupil is assegmented shown on the shape) in order to make further analysis. From the process of use of [18] to select the possible centreillumination in the light Fig. rather 5. thanThis segmented image thegives pupil wrong boundary. normalization, the segmented image of the eye will give the coordinates first. The method consistEven of detection though the on pupil pupil boundary boundary is not accurately because detected value radius pupil and the iris. This image will be cropped threshold followed by checking if thein thissegmentation segmentation process,on pupil these is images segmented still can be base on the value of iris radius, so that the unwanted area will accepted since the incidence of cholesterol normally occur selected points (by threshold) correspond on the illumination light rather than be removed (e.g. sclera and limbic). Therefore only the from limbic up to pupil which is 30 percents from overall to a local minimum in their immediate segmented the pupil boundary. intended area can be analyzed. According to [8], arcus senilis normalization image. Therefore for this kind of segmentation neighbourhood these points serve as the is described as a yellowish-white ring around the cornea that is possible centre coordinates for the iris.we consider the correct segmentation of iris boundary rather than pupilEven boundarythough segmentation.the pupil boundary More examples is not for this separated from the limbus by a clear zone 0.3 to 1 mm in These radius values were set manually, processaccurately will show indetected the result’s in section. this segmentation width. Normally the area of white ring (Arcus Senilis), occurs (Table 1) shows the rmin and rmax varies Anotherprocess, example these as imagesshows in still the canFig. be6, failaccepted to determine from the sclera/iris up to 20 to 30 percents toward to pupil, so from 75 to 250 where the ranges for iristhe edgesince of the pupil incidence but it detect of cholesterol the edge normallyof the impurity this is the only the main area that have to be analyzed. The between 90 to 150 pixels and for pupilillumination occur fromlight. Thislimbic will up effect to pupilto the whichquality isof the other reason to normalize is to make the analysis process radius are 28 to 75 pixels. The input forsegmentation 30 percents eyes image, from cause overall to imperfectly normalization to detect iris become easier rather than to examine the eye in circular shape. this function is the image to be segmentedand pupilimage. boundary Therefore region of thefor eyes. this kind of In rectangular shape analyze can be done either from top to and the input parameters in this function segmentation we consider the correct including rmin and rmax (the minimum segmentation of iris boundary rather and maximum values of the iris radius). than pupil boundary segmentation. More The range of radius values to search for

ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 33 Journal of Telecommunication, Electronic and Computer Engineering

examples for this process will show in the V. NORMALIZATION result’s section. After the iris is localized, the next step Another example as shows in the Fig. is normalization (iris enrolment). It is a 6, fail to determine the edge of pupil process after localization (segmentation) but it detect the edge of the impurity intends to change the iris region to the illumination light. This will effect to the fixed form (circle shape) in order to make quality of the segmentation eyes image, further analysis. From the process of cause to imperfectly to detect iris and normalization, the segmented image of pupil boundary region of the eyes. the eye will give the value radius pupil and the iris. This image will be cropped neighbourhood these points serve as the possible centre base on the value of iris radius, so that coordinates for the iris. These radius values were set manually, the unwanted area will be removed (e.g. (Table 1) shows the rmin and rmax varies from 75 to 250 sclera and limbic). Therefore only the where the ranges for iris between 90 to 150 pixels and for intended area can be analyzed. According neighbourhoodpupil radiusthese arepoints 28 to se75rve pixels. as Thethe inputpossible for this centre function is coordinatesthe for image the iris. to Thesebe segmented radius valuesand the were input set parameters manually, in this to [8], arcus senilis is described as a (Table 1) showsfunction the including rmin and rmin rmax and varies rmax from (the 75minimum to 250 and yellowish-white ring around the cornea where the maximumranges for values iris ofbetween the iris radius).The90 to 150 rangepixels of andradius for values that is separated from the limbus by a clear Fig. 5: An example where segmentation pupil radiusto are search 28 tofor 75 was pixels. set manually, The input depending for this functionon the database is Fig. 5: An example where segmentation fails. The segmentation is failing zone 0.3 to 1 mm in width. Normally the to detectionfails. correctly The segmentation the edges of the pupil is border, failing but itto segmented detection the image used.to be Forsegmented the CASIA and database, the input rmin parameters is set to 55 in pixels this and rmax is set to 160 pixels. The sample of Arcus Senilis eye illuminationcorrectly light. the edges of the pupil border, but it area of white ring (Arcus Senilis), occurs function including rmin and rmax (the minimum and (arcus7.bmp) is downloading from (Mediscan, 2000), for this segmented illumination light. from the sclera/iris up to 20 to 30 percents maximum valuesimage the of rminthe iris is set radius).The to 70 pixels range and rmix of radius is set to values 260 pixels. But luckily the significant area of white ring (Arcus Senilis) toward to pupil, so this is the only the to search for was set manually, depending on the database lay at the boundary of sclera or iris up to the pupil, so as long Table 1 shows the others sample run for segmentation process. Fig. 5:But An example luckily where the segmenta significanttion fails. The segmentationarea of is failing white main area that have to be analyzed. The used. For the CASIA database, rmin is set to 55 pixels and to detectionas the correctlysegmentation the edges done of thecorrectly pupil border, on the but iris it itsegmented can considered illuminationsucceed.ring light. ( ThisArcus segmentation Senilis) layimage at willthe beboundary crop base on of the other reason to normalize is to make the rmax is set to 160 pixels. TheImage sample of Arcusrmin Senilisrmax eye (arcus7.bmp) is downloading from (Mediscan, 2000), for this valuesclera of iris’s or radius. iris up to the pupil, so as long analysis process become easier rather Iris1.bmp (CASIA) 75 260 But luckily the significant area of white ring (Arcus Senilis) image the rmin is set to 70Iris2.bmp pixels (CASIA)and rmix is set to75 260 pixels. 260 as the segmentation done correctly on than to examine the eye in circular Table 1 shows the others sampleIris3.bmp run (CASIA) for segmentation75 process. 260 lay at thethe boundary iris it canof sclera considered or iris up tosucceed. the pupil, soThis as long shape. In rectangular shape analyze can Iris4.bmp (CASIA) 75 260 as the segmentationsegmentation done image correctly will on thebe iriscrop it can base considered on be done either from top to bottom or Iris6.bmp (CASIA) 75 260 succeed. This segmentation image will be crop base on the Image c_mo1.bmp rmin rmax 50 83 the value of iris’s radius. from bottom to top. (J. Daugman, 2004) c_mo2.bmp 75 260 value of iris’s radius. Iris1.bmp (CASIA)normal1.bmp 75 75 260 260 describes details on algorithms used in Iris2.bmpArcus1.bmp (CASIA) (Medical Web)75 80 260 260 iris recognition. He has introduced the Iris3.bmpArcus2.bmp (CASIA) (Medical Web)75 80 260 260 Rubber Sheet Model that transforms the Iris4.bmpArcus3.bmp (CASIA) (Medical Web)75 80 260 260 bottom or from bottom to top. (J. Daugman, 2004) describes radius of the iris region ‘doughnut’ as a function of the angle Arcus4.bmp (Medical Web) 80 260 eye from circular shape into rectangular Iris6.bmp (CASIA) 75 260 details on algorithms used in iris recognition. He has θ. c_mo1.bmpArcus5.bmp (Medical Web) 50 8083 260 Fig. 6: Another example where segmentation fails to find the edges of the form and it is shown in Fig.7. This model Arcus6.bmp (Medical Web) 80 260 pupil border. A constant number of points are chosen along each radial c_mo2.bmp 75 260 introduced the Rubber Sheet Model that transforms the eye Arcus7.bmp (Medical Web) 80 260 remaps all point within the iris region to normal1.bmp 75 260 from circular shape into rectangular form and it is shown in line, so that a constant number of radial data points are taken, Arcus2.bmp (Medical Web) 80 260 The problem with this image is the existing of illumination Arcus1.bmp (Medical Web) 80 260 Fig.7.a pair This ofmodel polar remaps coordinates all point within (r, θ),the whereiris region θ to a irrespective of how narrow or wide the radius is at a particular Arcus2.bmp (Medical Web) 80 260 from camera (on top of the pupil) will cause inaccurate Arcus2.bmpArcus2.bmp (Medical Web) (Medical Web) 80 80 260 260 angle. The normalised pattern was created by backtracking detection of iris and pupil boundary. This process considerpair is of the polar angle coordinates [0, 2 π(r,] θand), where r is θ on is the the angle interval [0, 2π] and Arcus3.bmp (Medicalubiris1.bmp Web) (UBIRIS) 80 85 260 260 to find the Cartesian coordinates of data points from the radial failure if the purpose of this segmentation process is used forr is [0,on the1]. interval [0, 1]. Arcus4.bmp (Medicalubiris2.bmp Web) (UBIRIS) 80 85 260 260 Fig. 6: Another example where segmentation and angular position in the normalised pattern. From the Arcus5.bmp (Medicalubiris2.bmp Web) (UBIRIS) 80 85 260 260 Fig.biometric 6: Another iris example recognition, where segmenta but for tionthis fails system to find since the edgesthe image of the fails to find the edges of the pupil border. ‘doughnut’ iris region, normalisation produces a 2D array with Arcus6.bmp (Medical Web) 80 260 pupilonly border. need to be analyze in 30 percent from the limbic (border horizontal dimensions of angular resolution and vertical Arcus7.bmpTable 1: shows (Medical the example Web) of the images 80 and their 260rmin and rmax values. of iris) so this result can be used. dimensions of radial resolution Arcus2.bmp (Medical Web) 80 260 The output of this function will be the value of c and c , The Theproblem problem with this imagewith isthis the existingimage of isillumination the The normalisation process proved to be successful and Arcus2.bmp (Medical Web) 80 260 p i V. NORMALIZATION which is the value of [xc, yc, r] for the pupilary boundary andfrom cameraexisting (on of top illumination of the pupil) from will cameracause inaccurate (on some results are shown in Fig 9. This normalization process Arcus2.bmp (Medical Web) 80 260 After the iris is localized, the next step is normalization (iris the limbicubiris1.bmp / iris (UBIRIS)boundary and also 85the segmented 260 image. Thedetection top of ofiris the and pupil)pupil boundary. will cause This inaccurateprocess consider will transforms the segmented eye from the circular form to programubiris2.bmp processes (UBIRIS) the image from 85[17] database 260 as shown infailure enrolment). detectionif the purpose It is of a ofirisprocess this and segmentation after pupil localization boundary. process (segmentation) is This used for the rectangular shape. However normalisation output will be the Fig. 5. This image gives wrong detection on pupil intends to change the iris region to the fixed form (circle ubiris2.bmp (UBIRIS) 85 260 biometricprocess iris recognition, consider but failure for this systemif the sincepurpose the image cropped until 30 percents, from the bottom eye (sclera and iris boundary because segmentation on pupil is segmented on theonly shape) need toin beorder analyze to make in further30 percent analysis. from From the thelimbic process (border of Fig. 7: Daugman’sFig. 7: Daugman’s rubber sheet model. rubber sheet model. boundary) toward pupil. Table 1: showsillumination the example light of ratherthe images than and segmented their rmin the and pupil rmax boundary.values. of iris)normalization, ofso thisthis result segmentation the can segmented be used. image process of the iseye used will give for the Normalisation of two eye images of the same iris is shown

Even though the pupil boundary is not accurately detected valuebiometric radius pupil iris and recognition,the iris. This image but will for be croppedthis The remapping of the iris image I(x, in Fig 9. The pupil is smaller in the bottom image, however The outputin thisof thissegmentation function willprocess, be the these value images of c stilland ccan, be The remapping of the iris image I(x, y) from Cartesian p i basesystem on the value since V.of iristheN ORMALIZATIONradius, image so that only the unwantedneed to area be will the normalisation process is able to rescale the iris region so accepted since the incidence of cholesterol normally occur coordinatesy) from to Cartesianthe norma coordinateslized non-concentric to the polar which is the value of [xc, yc, r] for the pupilary boundary and be removed (e.g. sclera and limbic). Therefore only the that it has constant dimension. Note that rotational from limbic up to pupil which is 30 percents from overall Afteranalyze the iris is inlocalized, 30 percent the next stepfrom is normalizationthe limbic (irisrepresentationnormalized can be modellednon-concentric as: polar the limbic / iris boundary and also the segmented image. The intended area can be analyzed. According to [8], arcus senilis inconsistencies have not been accounted for by the normalization image. Therefore for this kind of segmentation (border of iris) so this result can be used. I(x(r,θ ), y(r,θ )) → I(r,θ ) (3) program processes the image from [17] database as shown in enrolment).is described It isas a yellowish-whiteprocess after ringlocalization around the (segmentation) cornea that is representation can be modelled as: normalisation process, and the two normalised patterns are the Fig. 5.we Thisconsider image the correctgives segmenwrongtation detection of iris boundaryon pupil rather intends to change the iris region to the fixed form (circleWith slightly misaligned in the horizontal (angular) direction. than pupil boundary segmentation. More examples for this separated from the limbus by a clear zone 0.3 to 1 mm in Rotational inconsistencies will be accounted for in the boundary becauseprocess willsegmentation show in the onresult’s pupil section. is segmented on the shape)width. in order Normally to make the area further of white analysis. ring (Arcus From Senilis), the process occurs of x(r,θ ) = (1− r)x (θ ) + rx (θ ) (4) matching stage. illumination lightAnother rather example than segmentedas shows in the the pupil Fig. 6,boundary. fail to determine normalization, from the sclera/iris the segmented up to 20 toimage 30 percents of the toward eye will to pupil, give sothe p l It is difficult to do analysis if the image is in the original Even thoughthe edge the pupilof pupil boundary but it detectis not theaccurately edge of detected the impurity value this radius is the pupilonly theand main the areairis. thThisat have image to be will analyzed. be cropped The y(r,θ ) = (1− r)y p (θ ) + ryl (θ ) (5) form therefore the image needs to be wrapped to transform the in this segmentationillumination light.process, This thesewill effectimages to thestill qualitycan beof the other reason to normalize is to make the analysis process base on the value of iris radius, so that the unwanted area willWhere I(x, y) is the iris region image, (x, y) are the original nature from circle to rectangular shape. This process only can accepted sincesegmentation the incidence eyes image, of cholesterol cause to imperfectly normally to occur detect iris become easier rather than to examine the eye in circular shape. be removed34 (e.g. sclera andISSN: limbic). 2180 -Therefore 1843 Vol. only 3 Cartesian the No. 2 coordinates, July-December (r, θ) are 2011 the corresponding normalised be achieved by doing the conversion polar to rectangular. from limbicand up pupil to boundarypupil which region is of 30 the percents eyes. from overall In rectangular shape analyze can be done either from top to intended area can be analyzed. According to [8], arcus senilispolar coordinates, and xp, yp and xl, yl are the coordinates of the However, the normalisation process was not able to normalization image. Therefore for this kind of segmentation is described as a yellowish-white ring around the cornea thatpupil is and iris boundaries along the θ direction. The perfectly reconstruct the same pattern from images with we consider the correct segmentation of iris boundary rather localization of the iris and the coordinate system is able to separated from the limbus by a clear zone 0.3 to 1 mm in varying amounts of pupil dilation, since deformation of the iris than pupil boundary segmentation. More examples for this achieve invariance to 2D position and size of the iris, and to results in small changes of its surface patterns. process will show in the result’s section. width. Normally the area of white ring (Arcus Senilis), occursthe dilation of the pupil within the iris. Another example as shows in the Fig. 6, fail to determine from the sclera/iris up to 20 to 30 percents toward to pupil, soThe normalization process is illustrated in Fig. 8. It is done the edge of pupil but it detect the edge of the impurity this is the only the main area that have to be analyzed. Theby taking the reference point from the centre of the pupil and illumination light. This will effect to the quality of the other reason to normalize is to make the analysis processradial vectors pass via the iris area. There are two important segmentation eyes image, cause to imperfectly to detect iris become easier rather than to examine the eye in circular shape.data points along each radial line which are radial resolution and pupil boundary region of the eyes. for radial line in pupil and angular resolution for radial line In rectangular shape analyze can be done either from toparound to the iris region. Since the pupil can be non-matching with the iris therefore it need to remap to rescale the points depending to the angle around the iris and pupil. This formula given by: 2 2 (6) r'= α β ± αβ −α − rI

With

2 2 (7) α = ox + o y

⎛ ⎛ o ⎞ ⎞ (8) β = ⎜π − ⎜ y ⎟ −θ ⎟ cos⎜ arctan⎜ ⎟ ⎟ ⎝ ⎝ ox ⎠ ⎠

The shift of centre of the pupil relative to the iris centre, this given by Ox ,Oy, and r’ is the distance between edge of pupil Fig. 8: Outline of the normalization process with radial resolution of 10 and edge of the iris at the angle θ around the region, and r’ is pixels, and angular resolution of 40 pixels. the radius of the iris. The remapping formula first gives the bottombottom or from or bottomfrom bottom to top. to (J. top. Daugman, (J. Daugman, 2004) 2004)describes describes radius radius of the of iris the region iris region ‘doughnut’ ‘doughnut’ as a function as a function of the of angle the angle detailsdetails on algorithmson algorithms used usedin iris in recognition.iris recognition. He hasHe hasθ. θ. introducedintroduced the Rubber the Rubber Sheet SheetModel M thatodel transforms that transforms the eye the eye A constant A constant number number of points of points are chosen are chosen along alongeach radialeach radial line, so that a constant number of radial data points are taken, from circularfrom circular shape shapeinto rectan into rectangular gularform andform it and is shownit is shown in in line, so that a constant number of radial data points are taken, irrespective of how narrow or wide the radius is at a particular Fig.7. Fig.7.This modelThis model remaps remaps all point all pointwithin within the iris the region iris region to a to a irrespective of how narrow or wide the radius is at a particular angle.angle. The normalised The normalised pattern patte wasrn created was created by backtracking by backtracking pair ofpair polar of coordinatespolar coordinates (r, θ), (r,where θ), where θ is the θ isangle the angle[0, 2π ][0, and 2π ] and to findto the find Cartesian the Cartesian coordinates coordinates of data of points data points from thefrom radial the radial r is on the interval [0, 1]. r is on the interval [0, 1]. and angularand angular position position in the in normalisedthe normalised pattern. pattern. From Fromthe the ‘doughnut’‘doughnut’ iris region, iris region, normalisation normalisation produces produces a 2D arraya 2D witharray with horizontalhorizontal dimensions dimensions of angular of angular resolution resolution and verticaland vertical dimensionsdimensions of radial of radialresolution resolution The normalisationThe normalisation process process proved proved to be tosuccessful be successful and and some someresults results are shown are shown in Fig in 9. Fig This 9. normalizationThis normalization process process will transformswill transforms the segmented the segmented eye from eye thefrom circular the circular form toform to bottom or from bottom to top. (J. Daugman, 2004) describes theradius rectangularthe of rectangularthe iris shape. region shape. However ‘doughnut’ However normalisation as normalisationa function output of theoutput will angle be will be cropped until 30 percents, from the bottom eye (sclera and iris details on algorithms used in iris recognition. He has θ. cropped until 30 percents, from the bottom eye (sclera and iris Fig. 7: Daugman’sFig. 7: Daugman’s rubber sheetrubber model. sheet model. boundary) toward pupil. introducedbottom or fromthe Rubberbottom Sheetto top. M (J.odel Daugman, that transforms 2004) describes the eye radius A boundary)constant of the iris number toward region of pupil.‘doughnut’ points are aschosen a function along of each the radialangle NormalisationNormalisation of two of eye two images eye images of the ofsame the irissame is irisshown is shown fromdetails circular onAutomated algorithms shape into Detecting usedrectan ingular Arcus iris form Senilis,recognition. and it Symptom is shownHe has forin Cholesterolθline,. so that Presence a constant Using number Iris Recognitionof radial data Algorithmpoints are taken, in Figin 9. Fig The 9. pupil The pupilis smaller is smaller in the inbottom the bottom image, image, however however TheFig.7.introduced remapping This themodel Rubber of remaps the Sheetiris all impointMageodel within I(x,that y)transformsthe fromiris regionCartesian the toeye a irrespective A constant of numberhow narrow of points or wide are the chosen radius along is at aeach particular radial The remapping of the iris image I(x, y) from Cartesianthe normalisationthe normalisation process process is able is toable rescale to rescale the iris the region iris region so so coordinates to the normalized non-concentric polar line,angle. so that The a normalisedconstant number pattern of was radial created data pointsby backtracking are taken, pairfrom of coordinatescircular polar coordinates shape tointo (r,therectan θ ), normawheregular lizedθform is the and non-concentricangle it is [0, shown 2π] and in polarthat thatit hasit constanthas constant dimension. dimension. Note Notethat thatrotational rotational representation can be modelled as: toirrespective find the Cartesian of how narrow coordinates or wide of thedata radius points is from at a particularthe radial rFig.7. is onrepresentation Thisthe interval model [0,remaps can 1]. be allmodelled point within as: the iris region to a inconsistenciesinconsistencies have havenot beennot beenaccounted accounted for byfor theby the θ θ → θ (3) andangle.irrespective angular The positionnormalised of how in pattethe narrow rnnormalised was createdor widepattern. by backtracking the From the pair of polarI(x (coordinatesr,I(),x(yr(,rθ,), ))(r,y(r θ,θ),I ))where(r→, )I ( θr ,isθ )the angle [0, 2π] and (3)normalisation process, and the two normalised patterns are to‘doughnut’ findnormalisation the Cartesianiris region, process,coordinates normalisation and of the data produces two points normalised afrom 2D array the patterns radial with are With With slightlyradius misalignedis at a particularin the horizontal angle. (angular) The direction. r is on the interval [0, 1]. andhorizontal angularslightly dimensions positionmisaligned inof the inangular normalisedthe horizontal resolution pattern. (angular) and From vertical direction.the Rotationalnormalised inconsistencies pattern will wasbe accounted created for inby the ‘doughnut’dimensionsRotational irisof radial region, inconsistencies resolution normalisation will produces be accounted a 2D array for with in the x(r,θ )x=(r(,1θ−) r=)(x1p−(rθ))x+ rx(θ )(θ+)rx (θ ) (4) (4)matchingbacktracking stage. to find the Cartesian p l l horizontalThematching normalisation dimensions stage. process of angular proved resolution to be successful and vertical and It is difficultIt is difficult to do toanalysis do analysis if the ifimage the image is in theis in original the original y(r,θ )y=(r(,1θ−) r=)(y1p−(θr))y+ ry(θl)(θ+)ry (θ ) (5) (5)dimensionssomecoordinates results of are radial of shown data resolution in points Fig 9. Thisfrom normalization the radial process p l form therefore the image needs to be wrapped to transform the willandThe transforms form angularnormalisation therefore the position segmentedthe process image in needsprovedeye the from to betonormalised the wrappedbe circularsuccessful to formtransform and to the WhereWhere I(x, y) I(x, is the y) isiris the region iris region image, image, (x, y) (x,are y)the are original the original nature from circle to rectangular shape. This process only can somethe rectangular natureresults fromare shape. showncircle However toin rectangularFig 9. normalisation This shape. normalization This output process processwill onlybe can CartesianCartesian coordinates, coordinates, (r, θ) (r,are θthe) are correspo the corresponding ndingnormalised normalised bepattern. achieved byFrom doing the the conversion‘doughnut’ polar iris to rectangular. region, Where I(x,y) is the iris region image, (x, willcropped transformsbe untilachieved 30 the percents, by segmented doing from the conversion eyethe bottomfrom thepolar eye circular (sclerato rectangular. formand iris to polar coordinates,polar coordinates, and x pand, yp andxp, y xp l,and yl are xl, theyl are coordinates the coordinates of the of thenormalisation However, the normalisation produces processa 2D was arraynot able to Fig. 7: Daugman’s rubber sheet model. theboundary) rectangularHowever, toward shape. pupil.the However normalisation normalisation process outputwas notwill beable to pupil pupilandy) areirisand theboundariesiris originalboundaries along Cartesian alongthe θthe direction. coordinates,θ direction. The Theperfectlywith horizontal reconstruct thedimensions same pattern of from angular images with croppedNormalisationperfectly until 30 reconstruct percents,of two eye from theimages thesame bottom of thepattern eyesame (sclerafrom iris is andimages shown iris with localization localization(r, θ)of the are of iris the the and iris corresponding the and coordinate the coordinate system normalised system is able is toable varyingto amounts of pupil dilation, since deformation of the iris Fig. 7: Daugman’s rubber sheet model. boundary)inresolution Figvarying 9. The toward amountspupiland pupil. is ofverticalsmaller pupil dilation,in thedimensions bottom since deformationimage, of however of the iris achieveThe remappingachieve invariance invariance of to the 2D toirisposition 2D im positionage and I(x, size and y) of size fromthe of iris, Cartesianthe and iris, to and resultsto in small changes of its surface patterns. polar coordinates, and xp, yp and xl, yl theradial normalisationresults resolution. in small process changes is ableof its to surface rescale patterns. the iris region so the dilationthe dilation of the ofpupil the withinpupil within the iris. the iris. Normalisation of two eye images of the same iris is shown coordinatesare theto coordinatesthe normalized of thenon-concentric pupil and polariris that it has constant dimension. Note that rotational The normalizationThe normalization process process is illustrated is illustrated in Fig. in 8. Fig. It is8. done It is donein Fig 9. The pupil is smaller in the bottom image, however Therepresentation remappingboundaries can of be themodelled alongiris im as: age the I(x, θ direction.y) from Cartesian The inconsistencies have not been accounted for by the by takingby taking the reference the reference point frompoint thefrom centre the centre of the ofpupil the andpupil andtheThe normalisation normalisation process processis able to rescaleproved the toiris be region so coordinatesI( x(rto,θ ), ythe(r ,θ ))norma→ I(lizedr,θ ) non-concentric polar(3) normalisation process, and the two normalised patterns are radial radialvectorslocalization vectors pass viapass theof via iristhe the area. irisiris Therearea.and Theretheare twocoordinate are important two important that it has constant dimension. Note that rotational representationWith can be modelled as: slightlysuccessful misaligned and insome the horizontalresults are(angular) shown direction. data pointsdatasystem points along alongiseach able radialeach to radial lineachieve which line which invarianceare radial are radialresolution to resolution2D inconsistencies have not been accounted for by the I(x(r,θ ), y(r,θ )) → I(r,θ ) (3) Rotationalin Fig 9. inconsistenciesThis normalization will be accountedprocess willfor in the for radialforposition radialline in line pupil andin pupiland size angular and of angular theresolution iris,resolution forand radial forto radialtheline linenormalisation process, and the two normalised patterns are With x(r,θ ) = (1− r)x p (θ ) + rx (θ ) (4) matchingtransforms stage. the segmented eye from the aroundaround the iris the region. iris region. Since Sincethe lpupil the pupilcan be can non-matching be non-matching slightly misaligned in the horizontal (angular) direction. dilation of the pupil within the iris. RotationalcircularIt is difficult forminconsistencies to doto analysisthe willrectangular if thebe imageaccounted isshape. in thefor original in the with thewith yiris( rthe,θ therefore) iris= (1 −thereforer) yitp (needθ ) +it ryneedtol (θremap ) to remapto rescale to rescale the points the(5) points θ = − θ + θ (4) form therefore the image needs to be wrapped to transform the dependingdependingx (tor, the) angleto( 1the raroundangle)x p (around the) irisrx thel and( iris) pupil. and Thispupil. formula This formula given givenmatchingHowever stage. normalisation output will Where I(x, y) is the iris region image, (x, y) are the original nature from circle to rectangular shape. This process only can by: by:The y(r,θ )normalization= (1− r)y (θ ) + ry process(θ ) is illustrated(5) beIt iscropped difficult tountil do analysis 30 percents, if the image from is in the original Cartesian coordinates, (r,p θ) are thel corresponding normalised(6) be achieved by doing the conversion polar to rectangular. in Fig. 8. It is done2 2 by2 taking2 the reference (6)form therefore the image needs to be wrapped to transform the polar coordinates,r'= αr'β= and±α βαβx ,± y − αβandα −x−r, Iyα are− r the coordinates of the bottom eye (sclera and iris boundary) Where I(x, y) is the irisp regionp limage,l I(x, y) are the original natureHowever, from circlethe tonormalisation rectangular shape.process This was process not onlyable canto pupil andpoint iris fromboundaries the centrealong the of θthe direction. pupil andThe CartesianWith coordinates, (r, θ) are the corresponding normalised beperfectlytoward achieved reconstructpupil. by doing thethe conversion same pattern polar tofrom rectangular. images with localizationWith of the iris and the coordinate system is able to polar coordinates,radial vectors and xp, y passp and x vial, yl arethe the iris coordinates area. There of the varyingHowever, amounts the ofnormalisation pupil dilation, processsince deformation was not ofable the iristo achieve invariance to 2D position and size of the iris, and to pupil andare iristwo boundaries important2 2 2along data2 the points θ direction. along each The(7) results in small changes of its surface patterns. α = oxα+=o y + (7)perfectlyNormalisation reconstruct ofthe twosame eyepattern images from images of with the dilation of the pupil withinox theo yiris. localizationradial of theline iris which and the are coordinate radial systemresolution is able for to varying amounts of pupil dilation, since deformation of the iris achieveThe normalization invariance to process 2D position is illustrated and size in of Fig. the 8. iris, It isand done to the same iris is shown in Fig 9. The radial line⎛ in pupil⎛ o ⎞and⎞ angular resolution(8) (8)results in small changes of its surface patterns. by taking the reference point⎛ fromy ⎛ theo y ⎞ centre⎞ of the pupil and pupil is smaller in the bottom image, the dilation ofβ the= cos pupilβ⎜π= −cos arctanwithin⎜π −⎜arctan the⎟ − iris.⎜θ ⎟ ⎟ −θ ⎟ for radial⎜ line⎜ around⎜ o ⎟ ⎜ the⎟ ⎟ iris⎟ region. Since radialThe vectorsnormalization pass ⎝via process the⎝ iris⎝ is x area.illustrated⎠ ⎝ o ⎠xThere⎠ in⎠ are Fig. two 8. importantIt is done however the normalisation process is bydata taking pointsthe the along pupilreference each can radialpoint be fromline non-matching whichthe centre are radialof the with resolutionpupil theand able to rescale the iris region so that it has for radial line in pupil and angular resolution for radial line radialThe vectorsshiftirisThe of shifttherefore centrepass of viacentre of the it ofpupiliris needthe area. repupillative toThere re remap lativeto theare to iristwo theto centre, important irisrescale centre, this thisconstant dimension. Note that rotational givenaround by the O iris,O ,region. and r’ Sinceis the thedistance pupil between can be non-matchingedge of pupil Fig. 8: OutlineFig. 8: Outline of the normalization of the normalization process pr withocess radial with resolution radial resolution of 10 of 10 data pointsgiventhe xalongby points yOx each,Oy ,depending andradial r’ isline the which distanceto the are anglebetweenradial resolutionaround edge of pupil with the iris therefore it need to remap to rescale the points pixels,inconsistencies andpixels, angular and angular resolution resolutionhave of 40 pixels.not of 40 beenpixels. accounted andfor radialedgeandthe ofedgeline theiris in ofiris andpupilthe at iristhe pupil.and atangle theangular Thisangleθ around resolution θformula around the region, thefor given region,radial and by:r’line and is r’ is thedependingaround radiusthe the radius toof iristhe the angleregion.of iris. the around The iris.Since remapping theThe the iris remapping pupiland formulapupil. can Thisformulabe firstnon-matching formula givesfirst given givesthe thefor by the normalisation process, and by:with the iris therefore it need to remap to rescale the points the two normalised patterns are slightly (6) depending to the angle around2 the iris and2 pupil. This formula given r'= α β ± αβ −α − rI misaligned in the horizontal (angular) by: With (6) direction. Rotational inconsistencies will = α β ± αβ 2 −α − 2 r' rI be accounted for in the matching stage. 2 2 (7) With With α = ox + o y It is difficult to do analysis if the image is 2 2 (7) α⎛= ox + o y⎛ o ⎞ ⎞ (8) β = cos⎜π − arctan⎜ y ⎟ −θ ⎟ in the original form therefore the image ⎜ ⎜ o ⎟ ⎟ ⎝ ⎝ x ⎠ ⎠ needs to be wrapped to transform the ⎛ ⎛ o ⎞ ⎞ (8) β = cos⎜π − arctan⎜ y ⎟ −θ ⎟ nature from circle to rectangular shape. The shift of centre⎜ of the pupil⎜ o ⎟ relative⎟ to the iris centre, this ⎝ ⎝ x ⎠ ⎠ This process only can be achieved by given by O ,O , and r’ is the distance between edge of pupil Fig. 8: Outline of the normalization process with radial resolution of 10 x y doing the conversion polar to rectangular. andThe edge shift of ofthe centre iris at of the the angle pupil θre aroundlative to the the region, iris centre, and r’this is pixels, and angular resolution of 40 pixels. However, the normalisation process giventhe radius byThe O ofx ,O shiftthey, andiris. r’ofThe iscentre theremapping distance of the formulabetween pupil firstedge relative gives of pupil the to Fig. 8: Outline of the normalization process with radial resolution of 10 and edge of the iris at the angle θ around the region, and r’ is pixels,was andnot angular able resolution to of 40perfectly pixels. reconstruct the iris centre, this given by Ox ,Oy, and r’ the radiusis theof the distance iris. The betweenremapping edgeformula of first pupil gives and the the same pattern from images with edge of the iris at the angle θ around the varying amounts of pupil dilation, since region, and r’ is the radius of the iris. The deformation of the iris results in small remapping formula first gives the radius changes of its surface patterns. of the iris region ‘doughnut’ as a function of the angle θ.

A constant number of points are chosen along each radial line, so that a constant number of radial data points are taken,

ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 35 bottom or from bottom to top. (J. Daugman, 2004) describes radius of the iris region ‘doughnut’ as a function of the angle details on algorithms used in iris recognition. He has θ. pupil. This can only be done by searching the centre point of introduced the Rubber Sheet Model that transforms the eye A constant number of points are chosen along each radial the pupil given by x and y axis as propose by [18]. Hough from circular shape into rectangular form and it is shown in line, so that a constant number of radial data points are taken, transform [18] is used to detect edge of the iris and pupil Fig.7. This model remaps all point within the iris region to a irrespective of how narrow or wide the radius is at a particular circle. angle. The normalised pattern was created by backtracking pair of polar coordinates (r, θ), where θ is the angle [0, 2π] and Next, the image has to be analyzed and this can only be to find the Cartesian coordinates of data points from the radial r is on the interval [0, 1]. done if it is transformed to normalized polar coordinates using and angular position in the normalised pattern. From the Rubber Model. Since the “sodium ring”, terminology given in ‘doughnut’ iris region, normalisation produces a 2D array with iridology, or arcus senilis for the greyish or whitish arc in iris horizontal dimensions of angular resolution and vertical is only available at the bottom of this coordinate, thus only dimensions of radial resolution 30% (Fig.11), of the iris part is considered in the The normalisation process proved to be successful and normalization. some results are shown in Fig 9. This normalization processFig 9: Illustration of the normalization process, normal and illness eye. will transforms the segmented eye from the circular form to Lastly, to determine whether the eye has the ring, histogram of the image has to be plotted so that the decidability can be the rectangular shape. However normalisation output will be cropped until 30 percents, from the bottom eye (sclera and iris determined using OTSU’s method. The algorithm assumes Fig. 7: Daugman’s rubber sheet model. boundary) toward pupil. the image contains two classes of pixels (e.g. foreground and Normalisation of two eye images of the same iris is shown background) and finds the optimum threshold separating the in Fig 9. The pupil is smaller in the bottom image, however two classes so that their combined spread (within-class

The remapping of the iris image I(x, y) from Cartesian the normalisation process is able to rescale the iris region so variance) is minimal. coordinates to the normalized non-concentric polar that it has constant dimension. Note that rotational representation can be modelled as: Fig. 10: Color and grey scale eye images using [14] database. inconsistencies have not been accounted for by the VII. RESULTS I(x(r,θ ), y(r,θ )) → I(r,θ ) (3) normalisation process, and the two normalised patterns are A clear view for this cholesterol detection system With slightly misaligned in the horizontal (angular) direction. Rotational inconsistencies will be accounted for in the demonstrate in Fig. 13 whereas shows the entirely process of θ = − θ + θ x(r, ) (1 r)x p ( ) rxl ( ) (4) matching stage. the cholesterol detection system using iris recognition and 100% y(r,θ ) = (1− r)y (θ ) + ry (θ ) (5) It is difficult to do analysis if the image is in the original image processing algorithm. This process comprises the p l form therefore the image needs to be wrapped to transform the 30% following actions: Where I(x, y) is the iris region image, (x, y) are the original nature from circle to rectangular shape. This process only can Cartesian coordinates, (r, θ) are the corresponding normalised be achieved by doing the conversion polar to rectangular. Fig. 11: Stages of normalization 100 percents and 30 percents display after a) Eye images acquire from database (CASIA, UBIRIS, polar coordinates, and xp, yp and xl, yl are the coordinates of the However,Journal ofthe Telecommunication, normalisation process Electronic was not and ableComputer transformationto Engineering from polar to rectangular. MMU and medical web) or from digital camera. pupil and iris boundaries along the θ direction. The perfectly reconstruct the same pattern from images with localization of the iris and the coordinate system is able to b) Process of pupil and iris localization and segmentation, varying amounts of pupil dilation, since deformation of the irisThe normalization process is used for converting the to classify the required region. achieve invariance to 2D position and size of the iris, and to results in small changes of its surface patterns. circular iris into rectangular form with fixed dimension as c) Attain normalization iris from circular shape to the dilation of the pupil within the iris. rectangular shape (normalisation) and The normalization process is illustrated in Fig. 8. It is done shown in Fig 12. We can see clearly the circle shape rectangular shape with full image. by taking the reference point from the centre of the pupil and (localisation)also the turn signs to inrectangular both pictures shape (normalisation)are labelled and d) Crop the normalization iris to 30% from full image, (as radial vectors pass via the iris area. There are two important also theto illustratesigns in both the pictures upper are eyelid, labelled pupil to illustrate and the shown in Fig.11). data points along each radial line which are radial resolution upperwhite eyelid, pupildot exitand whiteduring dot exitsegmentation during segmentation and and e) Analyze the normalization iris to get the histogram for radial line in pupil and angular resolution for radial line after normalisation.after normalisation. value. around the iris region. Since the pupil can be non-matching f) Using OTSU to calculate the optimum threshold to with the iris therefore it need to remap to rescale the points detect arcus senilis (Cholesterol presence). depending to the angle around the iris and pupil. This formula given g) Results “Sodium ring detected” or “not detected” will by: (6) be display in MATLAB window. 2 2 r'= α β ± αβ −α − rI With For this experiment the boundary value of threshold in the examined eye image is set to 139 (this was decide after run 50 2 2 (7) samples of the eye image), the average boundary of the illness α = ox + o y or detected eye problem with sign of the presence cholesterol ⎛ ⎛ o ⎞ ⎞ (8) deposited. If the threshold value fallen below this value the β = ⎜π − ⎜ y ⎟ −θ ⎟ cos⎜ arctan⎜ ⎟ ⎟ eye image is considered as normal (no existing white ring or ⎝ ⎝ ox ⎠ ⎠ cholesterol), but if the threshold value rise up beyond this Fig. 12: Stages of normalization. The shift of centre of the pupil relative to the iris centre, this Fig. 12: Stages of normalization. value (139), the subject or patient is detected a sign of the presence of cholesterol. The algorithm is written using given by Ox ,Oy, and r’ is the distance between edge of pupil Fig. 8: OutlineFig. 8: of Outline the normalization of the pr ocessnormalization with radial resolution process of 10 pixels, and angular resolution of 40 pixels. VI. CHOLESTEROL DETECTION SYSTEM MATLAB. The result is based on the value obtained in the and edge of the iris at the angle θ around the region, and r’ is with radial resolution of 10 pixels, and automated detecting cholesterol presence (ADCP). The output the radius of the iris. The remapping formula first gives the angular resolution of 40 pixels. TheVI. process startsCH OwithLES obtaTEiningRO numberL of normal eyes and illness eye images (arcus senilis). of algorithm will be either “sodium ring is detected” or “no The next step isD toE TisolateECTION the actual SYS iris TregionEM in digital eye sodium ring is detected” depend on the sampling eye images pupil. This can only be done by searching the centre point of image. The pupil.isolation This process can only needs be doneto be bydone searching to segment the centrethe pointprocessed of by the ADCP. By using this ADCP will determine The theprocess pupil startsgiven by with x and obtaining y axis as number propose by [18]. Hough outer boundarythe pupil for thegiven iris by and x andthe innery axis boundary as propose for by the [18]. either Hough someone have the symptom of the cholesterol presence transform [18] is used to detect edge of the iris and pupil of pupil.normaltransform This caneyes [18] only and isbe useddone illness toby searchingdetect eye edge imagesthe centreof the point iris ofand pupil circle. (arcusthecircle. pupilsenilis given). by x and y axis as propose by [18]. Hough transformNext,Next, [18] thethe imageisimage used hastohas detect toto bebe edge analyzedanalyzed of the and andiris thisandthis canpupilcan only only bebe Thecircle. donenextdone if ifstep it it is is transformed transformedis to isolate to to normalized normalizedthe actual polar polar iris coordinates coordinates using using Rubber Model. Since the “sodium ring”, terminology given in regionNext,Rubber in thedigital Model. image eye hasSince toimage. bethe analyzed“sodium The andisolation ring”, this terminology can only be given in doneiridology,iridology, if it is transformed or or arcus arcus senilis senilis to normalized for for th thee greyishpolar greyish coordinates or or whitish whitish using arc arc in in iris iris process needs to be done to segment the Rubberisis onlyonly Model. availableavailable Since at theat the the“sodium bottombottom ring”, ofof terminologythisthis coordinate,coordinate, given thusinthus onlyonly outeriridology,30%30% boundary (Fig.11), (Fig.11),or arcus forsenilis ofof the theforthe iris th iriseiris greyishand partpart the or is whitishisinner consideredconsidered arc in iris inin thethe boundaryisnormalization. only available for the at pupil.the bottom This of thiscan coordinate, only be thus only Fig 9: Illustration of the normalization process, normalization. FigFig 9: 9: Illustration Illustration of of the the normaliza normalizationtion process, process, normal normal and and illness illness eye. eye. done30% byLastly, (Fig.11), searching to determineof the iriswhethercentre part pointthise eyeconsidered of has the the ring,in thehistogram normal and illness eye. normalization.Lastly, to determine whether the eye has the ring, histogram Fig 9: Illustration of the normalization process, normal and illness eye. pupilofof giventhe the image image by has hasx andto to be be yplotted plotted axis so asso that thatpropose the the decidability decidability can can be be Lastly, to determine whether the eye has the ring, histogram ofdetermined determinedthe image has usingusing to be OTSU’s OTSU’splotted so method.method. that the decidability TheThe algorithmalgorithm can be assumes assumes by [18]. Hough transform [18] is used to detectdeterminedthethe edgeimage image usingof contains contains the OTSU’s iris two two and method.classes classes pupil of ofThe circle.pixels pixels algorithm (e.g. (e.g. foreground assumesforeground and and thebackground)background) image contains andand two findsfinds classes thethe of optimumoptimum pixels (e.g. thresholdthreshold foreground separatingseparating and thethe background)twotwo classesclasses and sofindsso thatthat the theiroptimumtheir coco mbinedthresholdmbined spreadseparatingspread (within-class(within-class the Next,two theclasses image so that has their to becombined analyzed spread and (within-class variance)variance) is is minimal. minimal. thisvariance) can only is minimal. be done if it is transformed Fig.Fig. 10: Fig.10: Color Color 10: and Colorand grey grey scaleand scale eye greyeye images images scale using using eye [14] [14] images database. database. using Fig. 10: Color and grey [14]scale eyedatabase. images using [14] database. to normalized polar coordinatesVII.VII. RESULTSRESULTS using Rubber Model. SinceVII. theRESULTS “sodium ring”, AA clearclear viewview forfor thisthis cholesterolcholesterol detectiondetection systemsystem terminologyA clear viewgiven for in this iridology, cholesterol or detectionarcus system demonstratedemonstrate in in Fig. Fig. 13 13 wherea whereass shows shows the the entirely entirely process process of of senilisdemonstrate for thein Fig. greyish 13 wherea sor shows whitish the entirely arc process of thethethe cholesterol cholesterolcholesterol detection detectiondetection system systemsystem using usingusingiris recognition irisiris recognitionrecognition and andand 100% 100% 100% in irisimageimageimage is processing only processingprocessing available algorithm. algorithm.algorithm. at This the This Thisprocess bottom processprocess comprises of comprisescomprises the thethe this coordinate, thus only 30% (Fig.11), 30% 30% followingfollowing actions: actions: 30% following actions: of the iris part is considered in the Fig. 11: 11: Stages Stages of normalization of normalization 100 percents and 100 30 percents percents display after a) Eye images acquire from database (CASIA, UBIRIS, Fig.Fig. 11: 11: Stages Stages of of normalization normalization 100 100 pe percentsrcents and and 30 30 percents percents display display after afternormalization. a)a) Eye Eye imagesimages acquireacquire fromfrom databasedatabase (CASIA,(CASIA, UBIRIS,UBIRIS, transformationand 30 percents from polar to display rectangular. after transformation MMU and medical web) or from digital camera. transformationtransformation from from polar polar to to rectangular. rectangular. MMUMMU and and medical medical web) web) or or from from digital digital camera. camera. from polar to rectangular. b)b) Process Process of pupilof pupil and andiris localizationiris localization and segmentation, and segmentation, The normalization process is used for converting theLastly, b)toto classifydetermineProcess the of required pupil whether and region. iris the localization eye has and segmentation, TheThe normalizationnormalization processprocess isis usedused forfor convertingconverting thethe to classify the required region. circular iris into rectangular form with fixed dimension asthe ring,c) Attain histogramto classify normalization the of required the iris image region.from circular has to shape to circularcircularThe irisiris intonormalizationinto rectangularrectangular form formprocess withwith fixedfixed is dimensiondimensionused asas c) Attain normalization iris from circular shape to shown in Fig 12. We can see clearly the circle shapebe plottedc)rectangular Attain so that shapenormalization the with decidability full image. iris from can circularbe shape to shownshownfor inin FigconvertingFig 12.12. WeWe cancanthe see seecircular clearlyclearly thetheiris circle circleinto shape shape rectangularrectangular shape shape with with full full image. image. (localisation) turn to rectangular shape (normalisation) anddetermined d) Crop theusing normalization OTSU’s iris method. to 30% from Thefull image, (as (localisation)(localisation)alsorectangular the signs turnturn in to to formboth rectangularrectangular pictures with areshapeshape fixed labelled (normalisation)(normalisation) dimensionto illustrate theandand d)d)shown Crop Crop in the Fig.11).the normalization normalization iris iris to to 30% 30% from from full full image, image, (as (as algorithm assumes the image contains alsoalso theuppertheas signsshownsigns eyelid, inin pupil bothinboth and Fig picturespictures white 12. dot Weareare exit labelledlabelledcan during see segmentation toto clearly illustrateillustrate and thethe e) Analyzeshownshown thein in Fig.11). Fig.11).normalization iris to get the histogram two classesvalue. of pixels (e.g. foreground upperupperafter theeyelid, eyelid, normalisation. circle pupil pupil and andshape white white dot(localisation) dot exit exit during during segmentation segmentationturn to and and e)e) Analyze Analyze thethe normalizationnormalization irisiris toto getget thethe histogramhistogram f) Using OTSU to calculate the optimum threshold to after normalisation. and background)value.value. and finds the optimum after normalisation. detect arcus senilis (Cholesterol presence). f)f) Using Using OTSUOTSU toto calculatecalculate thethe optimumoptimum thresholdthreshold toto g) Resultsdetect “Sodium arcus senilis ring detected(Cholesterol” or “not presence). detected” will be displaydetect arcusin MATLAB senilis window.(Cholesterol presence). g) Results “Sodium ring detected” or “not detected” will 36 ISSN: 2180 - 1843 Vol. 3 No. 2 July-Decemberg) Results “Sodium 2011 ring detected” or “not detected” will For thisbebe experiment display display in in theMATLAB MATLAB boundary window. window.value of threshold in the examined eye image is set to 139 (this was decide after run 50 samplesForFor ofthis this the experiment experimenteye image), the the averageboundary boundary boundary value value ofof of thethreshold threshold illness in in the the orexamined examineddetected eye eye eye problem image image iswith is set set si to gnto 13 of139 9the (this (this presence was was decide decidecholesterol after after run run 50 50 deposited.samplessamples Ifof of thethe the thresholdeye eye image), image), value the the fallen average average below boundary boundary this value of of thethe the illness illness eyeoror imagedetected detected is considered eye eye problem problem as normal with with si(nosigngn existingof of the the presencewhitepresence ring cholesterol cholesterolor cholesterol), but if the threshold value rise up beyond this deposited.deposited. IfIf thethe thresholdthreshold valuevalue fallenfallen belowbelow thisthis valuevalue thethe Fig. 12: Stages of normalization. valueeye image(139), theis consideredsubject or patientas normal is detected (no existing a sign whiteof the ring or presenceeye image of cholesterol.is considered Th eas algorithmnormal (no is existingwritten usingwhite ring or cholesterol), but if the threshold value rise up beyond this VI. CHOLESTEROL DETECTION SYSTEM MATLAB.cholesterol), The resultbut if is thebased threshold on the valuevalue obtained rise up in beyondthe this Fig.Fig. 12: 12: Stages Stages of of normalization. normalization. value (139), the subject or patient is detected a sign of the The process starts with obtaining number of normal eyes automatedvalue (139), detecting the cholesterol subject or presence patient (ADCP). is detected The outputa sign of the presence of cholesterol. The algorithm is written using and illness eye images (arcus senilis). of presencealgorithm willof cholesterol.be either “sodium The ringalgorithm is detected” is orwritten “no using VI. CHOLESTEROL DETECTION SYSTEM The nextVI. stepCHOLESTEROL is to isolate the DETECTION actual iris regionSYSTEM in digital eye sodiumMATLAB.MATLAB. ring is Thedetected”The resultresult depend isis basedbased on the onon sampling thethe valuevalue eye obtained obtainedimages inin thethe automated detecting cholesterol presence (ADCP). The output TheTheimage. processprocess The startsisolationstarts withwith process obtaobta needsiningining to numbernumber be done of ofto normalsegmentnormal eyestheeyes processedautomated by the detecting ADCP. cholesterolBy using this presence ADCP will (ADCP). determine The output eitherof algorithm someone havewill thebe symptoeither m“sodium of the cholesterol ring is detected” presence or “no andand illness illnessouter boundaryeye eye images images for (arcus (arcusthe iris senilis). senilis). and the inner boundary for the of algorithm will be either “sodium ring is detected” or “no sodium ring is detected” depend on the sampling eye images TheThe next next step step is is to to isolate isolate the the actual actual iris iris region region in in digital digital eye eye sodium ring is detected” depend on the sampling eye images processed by the ADCP. By using this ADCP will determine image.image. The The isolation isolation process process needs needs to to be be done done to to segment segment the the processed by the ADCP. By using this ADCP will determine either someone have the symptom of the cholesterol presence outerouter boundaryboundary forfor thethe irisiris andand thethe innerinner boundaryboundary forfor thethe either someone have the symptom of the cholesterol presence Automated Detecting Arcus Senilis, Symptom for Cholesterol Presence Using Iris Recognition Algorithm threshold separating the two classes so (ADCP). The output of algorithm will that their combined spread (within-class be either “sodium ring is detected” or variance) is minimal. “no sodium ring is detected” depend on the sampling eye images processed by the ADCP. By using this ADCP will VII. RESULTS determine either someone have the symptom of the cholesterol presence A clear view for this cholesterol or not. The result however is display in detection system demonstrate in Fig. 13 command window, but yet this program whereas shows the entirely process of can be run and display using Graphic the cholesterol detection system using Useror not. Interface The result however (GUI), is display where in command MATLAB window, has but Another result for cholesterol presence detection in normal iris recognition and image processing toolyet this to programperform can beit. run and display using Graphic User eye shown as in Fig. 16 below, where the eye images in this algorithm. This process comprises the Interface (GUI), where MATLAB has tool to perform it. figure shown variety of process involve from original eye until following actions: analysis to determine cholesterol presence.

Database from CASIA, 100% Normalization a) Eye images acquire from database MMU, UBIRIS Localization Iris Crop Localization & pupil Iris & pupil Polar to Rec. (CASIA, UBIRIS, MMU and medical web) or from digital

or not. The30% Normalization result however is display in command window, but Another result for cholesterol presence detection in normal camera. Result Cholesterol Polar to Rec. Histogram OTSU threshold b) Process of pupil and iris yet this program can be run and display usingpresence Graphic or not User eye shown as in Fig. 16 below, where the eye images in this InterfaceFig. 13:(GUI), Overall where system MATLAB for Cholesterol has Detection. tool to perform it. localization and segmentation, to Fig. 13: Overall system for Cholesterol figure shown variety of process involve from original eye until Fig 15 shows the Detectionhistogram, threshold values and the analysis to determine cholesterol presence. classify the required region. statement about the condition of the normal and Arcus Senilis c) Attain normalization iris from iris, respectively. From the histogram and OTSU’s method, Fig 15 shows the histogram, threshold Databasethe decidability from CASIA, or threshold value to distinguish100% Normalization between circular shape to rectangular shape MMU, UBIRIS Localization Iris Crop Localization & pupil Iris & pupil Polar to Rec. with full image. valuesnormal eyes and and eyesthe with statement Arcus Lipids is foundabout to be the139. conditionThis value is ofdetermined the normal after testing and 30 Arcusimages ofSenilis normal d) Crop the normalization iris to eyes. If the cluster mean value is less than this threshold, this 30% from full image, (as shown in iris,means respectively. than the eye is normal From eye and ifthe it is abovehistogram 139, then 30% Normalization Result Cholesterol andthe eye PolarOTSU’s can to Rec. be detected method,Histogram as eye with the ArcusOTSU decidability threshold Lipids. or Fig.11). presence or not Fig. 16: Results from normal eye threshold value to distinguish between e) Analyze the normalization iris to Fig. 13: Overall system for Cholesterol Detection. normal eyes and eyes with Arcus Lipids is The output from this experiment shown in Fig. 16 indicate get the histogram value. Fig 15 shows the histogram, threshold values and thethe eye image have no symptom of cholesterol presence this f) Using OTSU to calculate the statementfound toabout be the 139. condition This valueof the normal is determined and Arcus Senilis shown in command window result “Sodium ring not exist” optimum threshold to detect arcus iris,after respectively. testing From30 images the histogram of normal and OTSU’s eyes. method, If and the threshold value give 114 which is below the threshold the cluster mean value is less than this value (139). This shows the iris image contain no symptom of senilis (Cholesterol presence). the decidability or threshold value to distinguish betweencholesterol presence. Another result from abnormal eye image normal eyes and eyes with Arcus Lipids is found to be 139. g) Results “Sodium ring detected” or threshold,Original eye this meansSegmentation than processthe eye is is shown in Fig.17 below. The process follows the same This value is determined after testing 30 images of normalmethod as describe in above procedure. For this image the “not detected” will be display in normal eye and if it is above 139, then eyes. If the cluster mean value is less than this threshold, thisthreshold value determine is 148 which is higher than the set the eye can be detected as eye with Arcus point 139 thus the result in command window display the MATLAB window. means than the eye is normal eye and if it is above 139, then Lipids. message “Sodium ring had been detected”, this indicate the the eye can be detected as eye with Arcus Lipids. eye encompassFig. 16: ofResults cholesterol from normal presence eye symptom. For this experiment the boundary value 30 percents normalization OTSU output of threshold in the examined eye image The output from this experiment shown in Fig. 16 indicate the eye image have no symptom of cholesterol presence this is set to 139 (this was decide after run 50 Fig. 14: Stages of localization with eye image ‘Arcus1.bmp’ from the samples of the eye image), the average medical web, (Clock wise from top left) original colour eye image shown in command window result “Sodium ring not exist” localization, the iris and pupil detected correctly. (Top right) Gray eye image and the threshold value give 114 which is below the threshold boundary of the illness or detected localization (Bottom left) 30 percents display from polar to rectangular of value (139). This shows the iris image contain no symptom of localization eye with enhancement. (Bottom right) OTSU threshold value for eye problem with sign of the presence this eye is 144. cholesterol presence. Another result from abnormal eye image cholesterol deposited. If the threshold Original eye Segmentation process is shown in Fig.17 below. The process follows the same value fallen below this value the eye image method as describe in above procedure. For this image the threshold value determine is 148 which is higher than the set is considered as normal (no existing white point 139 thus the result in command window display the ring or cholesterol), but if the threshold message “Sodium ring had been detected”, this indicate the value rise up beyond this value (139), the eye encompass of cholesterol presence symptom. subject or patient is detected a sign of the 30 percents normalization Histogram output OTSU Finaloutput result presence of cholesterol. The algorithm Fig.Fig. 15: 14:Results Stages from eye of with localization “sodium ring’ withor Arcus eye senilis i.e. Fig. 17: Results from normal eye is written using MATLAB. The result ‘arcus1.bmp’:Fig.image 14: Stages ‘Arcus1.bmp’ histogram, of localization threshold vafrom withlue and eye thestatement image medical of ‘Arcus1.bmp’ iris condition web, from the is based on the value obtained in the medical web, (Clock wise from top left) original colour eye image localization,(Clock the wiseiris and from pupil detect toped left) correctly. original (Top right) colour Gray eye image automated detecting cholesterol presence localizationeye image(Bottom localization,left) 30 percents di thesplay iris from and polar pupil to rectangular of localization eye with enhancement. (Bottom right) OTSU threshold value for this eye is 144.

ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 37

Histogram output Final result

Fig. 15: Results from eye with “sodium ring’ or Arcus senilis i.e. Fig. 17: Results from normal eye ‘arcus1.bmp’: histogram, threshold value and statement of iris condition or not. The result however is display in command window, but Another result for cholesterol presence detection in normal yet this program can be run and display using Graphic User eye shown as in Fig. 16 below, where the eye images in this Interface (GUI), where MATLAB has tool to perform it. figure shown variety of process involve from original eye until analysis to determine cholesterol presence.

Database from CASIA, 100% Normalization MMU, UBIRIS Localization Iris Crop Localization or not. The result however is display& pupil in commandIris & pupil window,Polar to Rec. but Another result for cholesterol presence detection in normal yet this program can be run and display using Graphic User eye shown as in Fig. 16 below, where the eye images in this

Interface (GUI), where MATLAB has tool to perform it. figure shown variety of process involve from original eye until

30% Normalization Result Cholesterol Polar to Rec. Histogram OTSU threshold analysis to determine cholesterol presence. presence or not Fig. 13: Overall system for Cholesterol Detection.

Database from CASIA, Fig 15 shows the histogram,100% Normalization threshold values and the MMU, UBIRIS Localization Iris Crop Localization statement& pupil about theIris condition& pupil ofPolar the to Rec.normal and Arcus Senilis iris, respectively. From the histogram and OTSU’s method, the decidability or threshold value to distinguish between normal eyes and eyes with Arcus Lipids is found to be 139. 30% Normalization Result Cholesterol Polar to Rec.This valueHistogram is determinedOTSU thresholdafter testing 30 images of normal presence or not eyes. If the cluster mean value is less than this threshold, this Fig. 13: Overall system for Cholesterol Detection. means than the eye is normal eye and if it is above 139, then Fig 15 showsthe eyethe can histogram, be detected asthreshold eye with Arcusvalues Lipids and. the Fig. 16: Results from normal eye statement about the condition of the normal and Arcus Senilis iris, respectively. From the histogram and OTSU’s method, The output from this experiment shown in Fig. 16 indicate the decidability or threshold value to distinguish between the eye image have no symptom of cholesterol presence this normal eyes and eyes with Arcus Lipids is found to be 139. shown in command window result “Sodium ring not exist” and the threshold value give 114 which is below the threshold This value is determined after testing 30 images of normal value (139). This shows the iris image contain no symptom of eyes. If the cluster mean value is less than this threshold, this cholesterol presence. Another result from abnormal eye image means than the eyeOriginal is normal eye eye and if it isSegmentation above 139, process then is shown in Fig.17 below. The process follows the same the eye can be detected as eye with Arcus Lipids. method as describe in above procedure. For this image the Fig. 16:threshold Results from value normal determine eye is 148 which is higher than the set point 139 thus the result in command window display the The output from this experiment shown in Fig. 16 indicate Journal of Telecommunication, Electronic and Computermessage Engineering “Sodium ring had been detected”, this indicate the the eye eyeimage encompass have no of cholesterolsymptom presenceof cholesterol symptom. presence this 30 percents normalization shown in command window result “Sodium ring not exist” OTSU output and the threshold value give 114 which is below the threshold detected correctly. (Top right) Gray eye imagevalue (139).For Thisthis shows image the iris the image threshold contain no symptomvalue of Fig. localization14: Stages of localization (Bottom with left) eye image 30 ‘Arcus1.bmp’percents display fromcholesterol the determine presence. is Another 148 which result fromis higher abnormal than eye the image medical fromweb, (Clockpolar wiseto rectangular from top left) oforiginal localization colour eye eye image Original eye localization, the iris and pupilSegmentation detected correctly. process (Top right) Gray eye isimage shown set in point Fig.17 139 below. thus Thethe resultprocess infollows command the same with enhancement. (Bottom right) OTSU localization (Bottom left) 30 percents display from polar to rectangularmethod of windowas describe display in above theprocedure. message For this“Sodium image the localization eyethreshold with enhancement. value (Bottom for right)this OTSUeye is threshold 144. value for thresholdring value had determine been detected”, is 148 which this is higherindicate than the the set this eye is 144. point 139 thus the result in command window display the messageeye “Sodium encompass ring had ofbeen cholesterol detected”, this presence indicate the eye encompasssymptom. of cholesterol presence symptom. 30 percents normalization OTSU output

Fig. 14: Stages of localization with eye image ‘Arcus1.bmp’ from the medical web, (Clock wise from top left) original colour eye image Histogram output Final result localization, the iris and pupil detected correctly. (Top right) Gray eye image localization (BottomFig. left) Fig.15: 30 Results percents15: Results from di splayeye from withfrom “sodiumeyepolar with to ring’ rectangular “sodium or Arcus of senilisring’ i.e. Fig. 17: Results from normal eye localization eye with‘arcus1.bmp’: enhancement.or Arcus histogram, (B senilisottom threshold right) i.e. va ‘arcus1.bmp’:OTSUlue and statementthreshold of value irishistogram, condition for this eye is 144. threshold value and statement of iris condition

Another result for cholesterol presence detection in normal eye shown as in Fig. 16 below, where the eye images or not. The result however is display in command window, but Another result for cholesterol presence detection in normal in this figure shown variety of process yet this program can be run and display using Graphic User eye shown as in Fig. 16 below, where the eye images in this involve from original eye until analysis to Interface (GUI), where MATLAB has tool to perform it. figure shown variety of process involve from original eye until determine cholesterol presence. Histogramanalysis outputto determine cholesterolFinal presence. result

Fig. 17: Results from normal eye Fig. 15: Results from eye with “sodium ring’ or Arcus senilis i.e. Fig. 17: Results from normal eye ‘arcus1.bmp’: histogram, threshold value and statement of iris condition Database from CASIA, 100% Normalization MMU, UBIRIS Localization Iris Crop Localization & pupil Iris & pupil Polar to Rec.

VIII. CONCLUSION

30% Normalization This work introduces a non-invasive Result Cholesterol Polar to Rec. Histogram OTSU threshold presence or not method and iris recognition to detect Fig. 13: Overall system for Cholesterol Detection. the presence of cholesterol known as Fig 15 shows the histogram, threshold values and the hyperlipidemia by the sign of existence statement about the condition of the normal and Arcus Senilis arcus senilis in iris pigmented. Similar iris, respectively. From the histogram and OTSU’s method, opinion support by iridology practitioner the decidability or threshold value to distinguish between call this symptom as sodium ring refer to normal eyes and eyes with Arcus Lipids is found to be 139. arcus senilis sign of cardio heart diseases This value is determined after testing 30 images of normal (CHD). The algorithm had been tested eyes. If the cluster mean value is less than this threshold, this on more than 50 samples of normal and means than the eye is normal eye and if it is above 139, then abnormal eye images; it can be conclude the eye can be detected as eye with Arcus Lipids. that the threshold boundary of the normal Fig. 16: Results Fig.from normal16: Results eye from normal eye and problem eye is about 139. The outputThe output from this from experiment this experiment shown in Fig. shown 16 indicate the eyein image Fig. have 16 noindicate symptom the of cholesteroleye image presence have thisThe entire process of detecting cholesterol shown noin commandsymptom window of resultcholesterol “Sodium ringpresence not exist” presence using automated program and the threshold value give 114 which is below the threshold this shown in command window (ADCP), developed using MATLAB value (139). This shows the iris image contain no symptom ofcoding refer to Mr. Libor Masek’s work. cholesterolresult presence. “Sodium Another ring result not from exist” abnormal and eye the image threshold value give 114 which is below However this algorithm uses only a single Original eye Segmentation process is shown in Fig.17 below. The process follows the same methodthe as describethreshold in above value procedure. (139). ForThis this shows image themethod to determine the cholesterol sign thresholdthe value iris determine image iscontain 148 which no is symptomhigher than theof seti.e the arcus senilis which is using OTSU point 139cholesterol thus the resultpresence. in command Another window result display from themethod with histogram analysis. messageabnormal “Sodium ringeye had image been isdetected”, shown this in indicateFig.17 the eye encompassbelow. ofThe cholesterol process presence follows symptom. the same Other method can be used for determines 30 percents normalization the arcus senilis region such as using OTSU output method as describe in above procedure.

Fig. 14: Stages of localization with eye image ‘Arcus1.bmp’ from the medical web, (Clock wise from top left) original colour eye image 38 ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 localization, the iris and pupil detected correctly. (Top right) Gray eye image localization (Bottom left) 30 percents display from polar to rectangular of localization eye with enhancement. (Bottom right) OTSU threshold value for this eye is 144.

Histogram output Final result

Fig. 15: Results from eye with “sodium ring’ or Arcus senilis i.e. Fig. 17: Results from normal eye ‘arcus1.bmp’: histogram, threshold value and statement of iris condition Automated Detecting Arcus Senilis, Symptom for Cholesterol Presence Using Iris Recognition Algorithm neural network or sampling method for [8] F. L. Urbano, “Ocular Signs of detecting either normal or diseases eye Hyperlipidemia,” Hospital Physician, (arcus Senilis). no. November, pp. 51-54, 2001. Secondly the improvement can be done [9] N. Haq, M. D. Fox, A. Garton, and R. B. Northrop, “Mid Infrared Spectroscopic such as using graphic user interface (GUI) Absorption and Whole Blood for execute and displaying the result. This Cholesterol,” Blood, vol. i, pp. 33-34, program also can be used to determine 1991. the eye problem due to other type of eye diseases such as , , [10] D. Skin and C. Testing, “Issues in diabetic, tumour etcetera. Emerging Health Technologies,” Archives des Maladies du Coeur et des Vaisseaux, no. 34, 2002.

ACKNOWLEDGMENT [11] J.-Y. Um et.al., “Novel approach of molecular genetic understanding of The deepest gratitude and thanks to iridology: relationship between iris Universiti Teknikal Malaysia Melaka constitution and angiotensin converting (UTeM) and Faculty Electronic and enzyme gene polymorphism.,” The Computer Engineering (FKEKK) for American journal of Chinese medicine, supporting this publication. Thanks are vol. 33, no. 3, pp. 501-5, Jan. 2005. also due to Mr. Libor Masek for sharing [12] H. Z. Pomerantz, “The relationship MATLAB code of Daugman’s matching between coronary heart disease algorithm as public resource [5]. and the presence of certain physical characteristics.,” Canadian Medical Association Journal, vol. 86, pp. 57-60, REFERENCES Jan. 1962.

[1] O. Thefreedictionary, “Online [13] N. Useful and P. Harmful, “Not Useful dictionary,” Online dictionary, 1998. and Potentially Harmful,” vol. 118, pp. [Online]. Available: http://www. 120-121, 2000. thefreedictionary.com/iris. [14] UBIRIS Proenca, H. and Filipe, S. and [2] D. J. Pesek and P. D, “Iridology – An Santos, J. R. and Oliveira, and L. A. } Overview,” North, 2010. and Alexandre, “The [UBIRIS.v2]: A Database of Visible Wavelength Images [3] S. C. Reddy and U. R. Rao, “Ocular Captured On-The-Move and At-A- complications of adult rheumatoid Distance,” 2010. [Online]. Available: arthritis.,” Rheumatology international, http://iris.di.ubi.pt/ubiris2.html. vol. 16, no. 2, pp. 49-52, Jan. 1996. [15] UPOL, “UPOL,” 2003. [Online]. [4] G. Forbes, D. Feary, C. Savage, L. Available: http://phoenix.inf.upol.cz/ Nath, S. Church, and P. Lording, iris/download/. “Acute myeloid leukaemia (M6B: pure acute erythroid leukaemia) in [16] MMU, “MMU,” 2005. [Online]. a Thoroughbred foal.,” Australian Available: http://pesona.mmu.edu. veterinary journal, vol. 89, no. 7, pp. my/~ccteo/. 269-72, Jul. 2011. [17] CASIA, “CASIA,” 2003. [Online]. [5] L. Masek, “Recognition of Human Iris Available: http://www.cbsr.ia.ac.cn/ Patterns for Biometric Identification,” IrisDatabase.htm. Measurement, 2003. [18] T. A. Camus and R. Wildes, “Reliable [6] J. Daugman, “Iris Recognition,” and Fast Eye Finding in Close-up American Scientist, vol. 89, no. 4, p. Images,” Intelligence, 2002. 326, 2001. [19] W.-kin Kong and D. Zhang, “Detecting [7] J. Daugman, “How Iris Recognition and Reflection for Accurate Iris Works,” IEEE Transactions on Circuits Segmentation,” no. 852. and Systems for Video Technology, vol. 14, no. 1, pp. 21-30, Jan. 2004.

ISSN: 2180 - 1843 Vol. 3 No. 2 July-December 2011 39