sign in sign up
<<
Home , Aerial photography

Digitizing Aerial - Understanding Spatial Resolution

Trisalyn Nelson Mike Wulder Olaf Niemann M.Sc. Candidate Research Scientist Professor University of Victoria Pacific Forestry Centre University of Victoria (250) 721-7349 (250) 363-6090 (250) 721-7329 email: [email protected] [email protected] [email protected]

Abstract Accuracy, flexibility, and cost effectiveness are advantages of using digitized aerial as a data source. However, the relationship between the resolution of digital images, and the aerial photographs from which they were derived, must be addressed.

In the following paper we consider issues that impact the spatial resolution of photographs and digitized images, and suggest how users can optimize spatial resolution by selecting an appropriate scanning . Optimal scanning aperture can be chosen by considering the system resolution, the original ’s scale, and the desired size of the .

There are optimal scanning resolutions to use when digitizing aerial photographs. Optimizing spatial resolution will maximize the spatial information obtained from the original photographs without generating unnecessarily large file sizes.

Introduction Digitized aerial photography is increasingly the scanning aperture, thereby allowing being used as a data source for studying and important image information to be retained and managing environmental elements such as trees file sizes to be minimized. Choice of scanning (Niemann et al., 1999) and forests (Leckie et al., aperture is related to the resolution of the 1999; Niemann et al., 1999). The popularity of original photograph and/or the desired pixel size digitizing aerial photography is a result of the of the digital image. advantages of working with digitally formatted Although in this paper we only consider the data (Warner et al., 1996). For example, digital affect of scanning aperture on spatial resolution, imagery can be manipulated with relative ease it is recognized there are other impacting factors. allowing for atmospheric and geometric For example, systematic errors resulting from corrections, image enhancements, data shifts in the position of points within the pixel compression, and viewing options (Lillesand pattern reduces spatial resolution (Trinder, and Kiefer, 1994). As well, digital images 1987). As well, fitting homogenous square provide potential for semi-automated image to which are heterogeneous in analysis (Barbezat and Jacot, 1999; Gougeon, shape and size will adversely affect spatial 1999). One disadvantage of converting resolution (Fisher, 1997). Target and film photographs into digital imagery is that spatial contrast also affect the resolution of imagery; the resolution is reduced during the digitizing higher the contrast the better the potential to process (Warner et al., 1996). resolve unique features (Avery and Berlin, Spatial resolution is one factor which 1992). Additionally, scanner and input image governs the information content of an image. quality impact the spatial resolution of digitized Digitizing photography with the smallest imagery. Although we do not discuss it is in available scanning aperture reduces loss of this paper, it should be noted that other types of spatial resolution, however such an approach resolution such as reflectance and colour affect produces file sizes which are unnecessarily large the usability of digital imagery. and unwieldy to use. In this paper we suggest Depending on the anticipated use of digital techniques to optimize, rather than maximize, imagery, different approaches should be used to determine the scanning aperture required to the width of the lines (Lillesand and Kiefer, optimize spatial resolution. If digital imagery is 1994). The size of the smallest lines to be used for visual interpretation, a visual distinguishable on the film is the resolution of comparison of images scanned with varying the camera system (expressed in line pairs per sized should be made. The largest millimetre). scanning aperture which provides all required and film are the key information is best used to minimize file size. components which affect camera system Recent operational work by British Columbia’s resolution. The speed of a lens, or aperture Ministry of Environment Lands and Parks setting, controls the amount of light available at (MELP) suggest that when producing digital the film’s surface. As the aperture opens to imagery for visual interpretation, a scanning increase light at the film’s surface, image aperture between 10-14 microns will optimize sharpness decreases and vice versa. Therefore, the spatial resolution when digitizing film; and a the aperture setting is a compromise between scanning aperture of 36 microns is optimal when image sharpness and the amount of light digitizing prints (Fish, 2000). required for . Not all digital imagery is used for visual The film’s emulsion is composed of silver interpretation. Often it is used as input for halide grains, the size of which affects the image processing. Particularly when imagery is amount of light required to expose the film and used with semi-automated techniques, visual the graininess of the image. As silver halide interpretability becomes secondary to grains decrease in size image resolution maintaining the digital information required for improves, but more light is required for processing. In this paper our goal is to present exposure. Resolution may be sacrificed in order methods for selecting a scanning aperture which to achieve appropriate exposures (Avery and will optimize spatial resolution when converting Berlin, 1992 ). aerial photography film into digital imagery that is to be used for image processing. In the following paper we define spatial resolution, present methods for optimizing spatial resolution, and discuss example results of the spatial resolution optimization techniques used when scanning aerial photographs.

Spatial Resolution

Aerial Photography and Spatial Resolution Figure 1. Tri-bar resolution test chart (Avery and Berlin, 1992). Camera System Resolution There are many ways to define the spatial Measurements of camera system resolution resolution of a photograph or digital image. do not take into account the effects of converting Table 1 is a summary of the spatial resolution the image from a to photographic measurements discussed in this paper. The paper. Generally, the spatial resolution is spatial resolution of aerial photography is decreased during printing as the emulsion of the defined as the capability of a camera system to photographic paper has a lower spatial image spatial detail (Avery and Berlin, 1992 ). resolution than the emulsion of the film. In this paper we use Avery and Berlin’s (1992) Therefore, the camera system resolution is only definition of spatial resolution to define camera an accurate measure of the film, not the print, system resolution which is measured using a tri- resolution. bar resolution test chart (Figure 1). The tri-bar chart consists of consecutively smaller groups of three parallel lines separated by spaces equal to Table 1. Summary of the resolution measurements discussed in this paper. Adjacent resolution definitions for aerial photography and digital imagery are comparable. For example, the closest digital resolution measurement to camera system resolution is scanning aperture. Aerial Photography Digital Imagery Resolution Definition Units Resolution Definition Units Camera The capability of the camera line Scanning The pixel size used to sample microns System and film system to image pairs Aperture an image during digitizing. Resolution spatial detail (Avery and per mm As the size of the scanning Berlin, 1992). aperture decreases the spatial resolution theoretically increases. Ground The number of line pairs line Resolution which represent 1 m on the pairs ground (Avery and Berlin, per m 1992). Minimum The size of the smallest object cm or Pixel Size The ground area represented cm or m Detectable which can be detected. m by a specific size of scanning Object MDOS is equivalent to the aperture, when used in Size ground distance represented conjunction with a certain by one line of a line per mm scale of imagery. (MDOS) pair (Avery and Berlin, 1992). Effective The smallest object that can cm or Effective The smallest object that can cm or m Resolution be separated from the m Resolution be separated from the background. Similar to background. Similar to pixel MDOS, but also considers the size, but also considers the effects of contrast, effects of contrast, atmosphere, etc, (Hyppanen, atmosphere, etc. (Hyppanen, 1996). 1996).

Ground Resolution Ground resolution considers the effect of camera system resolution decreases, the ground scale and camera system resolution on the ability resolution also decreases. to identify ground objects (Avery and Berlin, 1992). Ground resolution is equivalent to the Minimum Detectable Object Size number of line pairs it takes to represent one Once the ground resolution is calculated, the metre on the ground and is calculated with the minimum detectable object size (MDOS) can be following equation: determined. The MDOS is the equivalent to the (Rs)(F) ground distance represented by one line of a line Rg = (1) per millimetre pair as measured by the tri-bar H chart (Avery and Berlin, 1992 ). To determine where: Rg = ground resolution in line pairs per the MDOS first calculate the ground resolution, metre then determine the ground distance represented R = camera system resolution in line s by a single line pair using the following pairs per millimetre equation: F = in millimetre H = height of camera above ground in 1 line pair width of a line pair in meters = (2) metres Rg The MDOS is the ground distance represented Photographic scale is a ratio of the focal length by one line. To calculate: and height of the camera above ground, therefore ground resolution is a function of MDOS = widthof onelinein meters = (3) photographic scale and camera system widthof a line pair in meters resolution. As the photographic scale and the 2 The width of one line in a line per millimetre Pixel Size pair is the equivalent to the MDOS. Pixel size is the most common way to report Theoretically, all objects equal in size to the the resolution of digital imagery and is MDOS can be identified. The MDOS is a approximately equivalent to the MDOS. For hypothetical measure of spatial resolution example, if a digital image has a pixel size equal (Rosenburg, 1971) and assumes that an object of to one metre, then the MDOS can be no greater size equal to the MDOS is fully recorded by a than one metre. The pixel size of a digital image single silver halide grain. This is unlikely as is based on the scanning aperture and the scale generally several grains record a portion of an of the original imagery, and is calculated with object and are used in conjunction with one the following equation: another to identify an object. Although the µm MDOS is a useful rule of thumb, it should be p = s × (5) 1,000,000 remembered that in practical terms the actual MDOS is likely smaller than reported. where: p = pixel size in metres s = scale of imagery Therefore, pixel size is the ground area Digital Imagery and Spatial Resolution represented by a scanning aperture when used in conjunction with a specific scale of imagery. Scanning Aperture The resolution of a digital image, generated Effective Resolution via the scanning of an aerial photograph, is The effective spatial resolution describes based on the scanning aperture (measured in what can actually be discerned on an image once microns) is used for digitizing. The scanning factors such as target contrast, atmospheric aperture is the size of the pixel used to sample condition, noise, and resampling have been an image. For example, an image digitized with accounted for (Hyppanen, 1996). The concept a scanning aperture of 14 microns is sampled at of effective resolution can be applied to both a rate of 714 pixels per centimetre. As the size photographs and digital imagery, however of the scanning aperture decreases the sampling generally refers to the latter. Although the rate increases, theoretically improving spatial spatial resolution of a digital image can never be resolution. Although dots per inch (dpi) is a greater than the pixel size it is often possible to term meant to express printer resolution, it is identify high contrast features at a sub-pixel commonly used to describe the sampling rate. level. For example, when working using To convert scanning aperture in microns to dpi imagery with a spatial resolution of 30 metres, use the following equation: roads, passing through forest, which are less than 30 metres wide are often visible due to their 25400 dpi = (4) high contrast. µm where µm = scanning aperture in microns Methods for Optimizing the Spatial Resolution of Digital Imagery Theoretically, if an image being scanned has Following is a discussion of two methods infinite resolution and the scanner is perfectly for optimizing the spatial resolution of a digital engineered the scanning aperture is a measure of image during scanning. Both methods require the digital image resolution. Although for most diapositives or film for input, and produce practical purposes the film’s spatial resolution is digital data for image processing. As well, both infinite (as it is almost always higher than the methods are based on controlling spatial spatial resolution available from even the resolution by selecting an appropriate scanning scanner’s smallest aperture setting), scanning aperture. The first optimization method is based errors should be expected to reduce spatial on maintaining the camera system resolution and resolution. Also, prints have a much lower the MDOS; the second method is based on spatial resolution than film and limit the generating a digital image with a pixel size potential spatial resolution of digital images. which optimizes the variance around an object of interest. 120 14 Calculating Scanning Aperture Based on 12 100 Camera System Resolution and MDOS 10 80 The aperture used for scanning may be 8 60 Ratio 6

40 Dots per mm selected so that the scanner system resolution 1:1 1:2.5 equals that of the camera system. Sampling 4 2 20 theories, such as the Kell factor, suggest that to 0 0 maintain spatial resolution the scanning aperture 254 64 36 25 20 16 13 12 10 9.1 8.2 7.5 must be set at 2.5 to 3 times the original Scanning Aperture (microns) resolution (Konecny et al., 1979). Sampling at a Figure 2. A comparison of scanning aperture and higher resolution than provided by the original line pairs per metre. If scanning at a 1:1 ratio an image is required to reduce systematic errors aperture of 19.5 microns should be used; if caused when the scanner signal does not line up implementing the Kell, factor, which has a ratio of 1:2.5, a scanning aperture of 7.95 microns should be precisely with the grains of the film. Choosing a used. scanning aperture with a 1:1 ratio could result in the loss of information if the sample is always Calculations for the MDOS are based on a taken at the bridge between two film grains. resolution equal to the camera system resolution. Therefore, if maintaining the information of This means that to maintain the MDOS an image every silver halide particle of the photograph is must be scanned using the same aperture as important the Kell factor should be required to maintain the camera system implemented. However, in many circumstances resolution. The MDOS is a measure of the this level of accuracy is not required as the maximum amount of information available from information available from the photograph is an image, therefore it makes sense that to greater than required for image analysis. Also, maintain the MDOS all the original information the large file sizes produced by applying the must be retained. Kell factor when scanning may prohibit certain Calculating the MDOS will not change the types of analysis. choice of scanning aperture; however, The camera system resolution is reported in knowledge of the MDOS may be important to lines pairs per millimetre, whereas scanner determine if an image provides enough resolution is reported in microns. By converting information to study a particular size of object. microns to pixels per millimetre, and line pairs For example, if the camera system resolution is per millimetre into lines per millimetre, the 50 line pairs per metre, the focal length of the resolution of the camera system and scanner can lens is 300 millimetres, and the flying be compared. In Figure 2 we present a height is 5000 metres, the ground resolution will comparison between the camera system equal three line pairs per metre. Therefore, one resolution measured in lines per millimetre and line pair equals 0.33 metres on the ground and the associated scanning aperture in microns. one line equals 0.17 metres on the ground. The The scanning aperture required to maintain both MDOS is 0.17 metres therefore, no object less the camera system resolution using a 1:1 ratio than 0.17 metres in size can be studied with this and the 1:2.5 ratio of the Kell factor are shown. imagery. Based on a hypothetical camera system resolution of 25 line pairs per millimetre, we Calculating Scanning Aperture Based on demonstrate that to maintain a camera system Size of Objects of Interest resolution with a 1:1 ratio a scanning aperture of Another way to determine the appropriate 19.5 microns is sufficient (Figure 2). However, scanning aperture is based on maximizing image if the Kell factor must be implemented a variance at the edge of an object of interest. scanning aperture of 7.94 should be used. Selecting a scanning aperture which results in high variance around an object is beneficial as it improves the accuracy of semi-automated feature extraction techniques. Preliminary required to maintain the MDOS. This suggests results suggest that pixels 0.17 the size of the that objects 0.5 metres in size cannot be object of interest result in high variance at an analyzed with this imagery. object’s boundary. Therefore, if the objects of In Table 3 we compare the scanning interest are trees with crown diameters of four aperture size for maintaining camera resolution metres, a pixel size no larger than 0.68 metres is and for maximizing image variance for a desirable. Although some trees with four metre particular object size. Unless the objects of crowns will be detected using a larger pixel size, interest are small, the size of the scanning the percentage of trees accurately located will aperture required to maximize pixel variance is decrease. larger than the scanning aperture required to Once the appropriate pixel size has been maintain the camera system resolution and determined, the scanning aperture can be MDOS. When digitizing to optimize the calculated using the following equation: variance of an object’s edge, the appropriate p scanning aperture can be selected without µm = (6) requiring camera calibration information, such s as camera system resolution. This is useful as The scanning aperture is a ratio of the scale of camera calibration information is not always the original photograph and the desired pixel available. It should be mentioned that size. optimizing the spatial resolution of a digital image based on maximizing the variance at an Results of Optimizing the Spatial object’s boundary is not always appropriate. For Resolution of Digital Imagery example, if an image is to be used by different In Table 2 we compare the scanning users for a variety of purposes or if several aperture suggested to maintain the camera objects of different size are being analyzed, it system resolution using both a 1:1 ratio and the may not be appropriate to select a scanning Kell factor ratio of 1:2.5. For all calculations in aperture based on object edge variance. Table 2 the scale is 1:15000. In Table 2 we demonstrate why implementation of the Kell factor may be impractical. Even if one has access to a scanner capable of digitizing with very small scanning apertures, doing so will generate large file sizes. Although data storage capacity and retrieval speeds are constantly increasing, file sizes of this magnitude are often impractical. For example, generating an requiring the use of several images having file sizes greater than a gigabyte will be difficult. Using large files, many processes could become onerous due to technical limitations and time constraints. Another factor to consider when scanning aerial photography is that if the object of interest is smaller than the MDOS, the imagery cannot provide enough spatial detail to analyze the object. The information provided by the photograph is not sufficient to detect objects smaller than the MDOS. In Table 3 we demonstrate that when the object size of interest is 0.5 metres, the scanning aperture required to maximize the variance of the object size of interest is smaller than the scanning aperture Table 2. Compares the dpi calculated to maintain the camera system resolution using a 1:1 ratio and the Kell factor of a 1:2.5 ratio. All image scales equal 1:15000. Camera Scanning Aperture File Size for Aerial Scanning Aperture File Size for Aerial System Required to Maintain Photograph Required to Maintain Photograph Resolution Camera Resolution (scanned at a 1:1 Camera Resolution (scanned at a 1:2.5) (line and MDOS with a ratio) 10 X 10 and MDOS with a ratio) 10 X 10 inches pairs/mm) 1:1 ratio (µm) inches in megabytes 1:2.5 ratio (µm) in megabytes 25 20.00 161 8.00 1008 50 10.00 645 4.00 4032 75 6.67 1452 2.67 9073 100 5.00 2581 2.00 16129

Table 3. Summary of optimal dpi suggestions. Note accuracy provided by the Kell factor is not that the dpi based on maximum variances is required. calculated with a pixel size equal to 0.17 the size of the object of interest. All calculations are based on a • Scanning an image with a smaller scanning 1:1 ratio. All image scales equal 1:15000 and 80 lppmm. aperture than is required to maintain the Scanning Aperture Object of Scanning Aperture camera system resolution and MDOS, will Required to Interest Required to theoretically improve the spatial resolution Maintain the Size (m) Maximize Image but no more useful information will be Minimum Variance Based on captured Detectable Object Object of Interest Size (µm) Size (µm) • 6.25 0.5 5.56 The scanning aperture suggested to maintain 6.25 1 11.11 the camera system resolution and MDOS 6.25 1.5 16.67 assumes the negative, not the print, is being 6.25 2 22.22 scanned. If scanning a print, the spatial 6.25 3 33.33 resolution will be reduced and a smaller 6.25 4 44.44 scanning aperture should be used. 6.25 5 55.56 6.25 6 66.67 • Using a scanning aperture which maximizes variance at the edge of a certain size of Conclusions object, will likely increase the aperture size Choosing an appropriate scanning aperture required for scanning. when digitizing aerial photography is a balancing act between information required and • If the objects of interest are smaller than the manageable file sizes. Following are a few rules MDOS the data can not provide sufficient which should guide the process of appropriate information for analysis. scanning aperture selection. Acknowledgments • It is best to optimize, rather than maximize, The authors would like to thank the spatial resolution when generating a digital following individuals and groups for their input image from an aerial photograph. and support: Don Leckie and David Hill from the Pacific Forestry Centre; Tom Doyle, Gordon • Regardless of the scanning aperture used, Fish, Allen Foster, Paul Quakenbush, Don data can never be created. Shelly, and Al Spring from the British Columbia’s Ministry of Environment, Lands and • Use the Kell factor cautiously. For most Parks; and John Wakeland from British general applications the Columbia’s Ministry of Forests. This project is supported in part by the GEOIDE project Dec #9 resolution digital imagery for forestry, Victoria, (www.geoide.ulaval.ca). B.C., Feb. 10-12, 1998, Natural Resources Canada, Canadian Forest Service, Pacific References Forestry Centre, 335-343. Avery, T. and Berlin, G. (1992), Fundamentals of Remote Sensing and Airphoto Interpretation., Lillesand, T. and Kiefer, R. (1994), Remote Maxwell Macmillan Canada, Toronto, 36-38. Sensing and Image Interpretation, John Wiley & Sons, Inc., USA, 49-144. Barbezat, V. and Jacot, J. (1999), The CLAPA Niemann, O., Adams, D. and Hay, G. (1999), project: Automated classification of forest with Automated tree crown identification using aerial photographs, International Forum on digital orthophoto mosaics, International Forum Automated Interpretation of High Spatial on Automated Interpretation of High Spatial Resolution Digital Imagery for Forestry, Resolution Digital Imagery for Forestry, Victoria, B.C., Feb. 10-12, 1998, Natural Victoria, B.C., Feb. 10-12, 1998, Natural Resources Canada, Canadian Forest Service, Resources Canada, Canadian Forest Service, Pacific Forestry Centre, 345-356. Pacific Forestry Centre, pp. 127-139.

Fish, G. (2000), British Columbia Ministry of Rosenburg, P. (1971), Resolution, delectability Environment, Lands and Parks. Personal and recognizability, Photogrammetric communication. Engineering. 37: 1255-1258.

Fisher, P. (1997), The pixel: a snare and a Trinder, J. (1987), Measurements of digitized delusion, International Journal of Remote hardcopy images, Photogrammertic Engineering Sensing. 18: 679-685. and Remote Sensing. 53: 315-321.

Gougeon, F. (1999), Automatic individual tree Warner, W., Graham, R. and Read, R. (1996), crown delineation using a valley-following Small Format Aerial Photography, American algorithm and a rule-based system, Automated Society for and Remote interpretation of high spatial resolution digital Sensing, Bethesda, Maryland, 348. imagery for forestry, Victoria, B.C., Feb. 10-12, 1998, Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, 11-24.

Hyppanen, H. (1996), Spatial autocorrelation and optimal spatial resolution of optical remote sensing data in boreal forest environment, International Journal of Remote Sensing. 17: 3441-3452.

Konecny, G., Bahr, H., Reil, W. and Schreiber, H. (1979), Use of spaceborne metric for cartographic applications, Institute of Photogrammetry and Engineering: Germany, 1- 165.

Leckie, D., Gillis, M., Gougeon, F., Lodin, M., Wakelin, J. and Yuan, X. (1999), Computer- assisted photointerpretation aids to forest inventory mapping: some possible approaches, Automated interpretation of high spatial