Examples of Filters Frequency domain Gaussian lowpass filter Gaussian highpass filter Spatial domain 1 Ideal Low Pass Filtering Low pass filter: Keep frequencies below a certain frequency Low pass filtering causes blurring After DFT, DC components and low frequency components are towards center May specify frequency cutoff as circle c Ideal Low Pass Filtering Multiply Image Fourier Transform F by some filter matrix m Shape of ILPF Frequency domain h(x,y) Spatial domain 4 Ideal low-pass filter (ILPF) 1 D(u,v) D0 H(u,v) 0 D(u,v) D0 D( u , v ) [( u M / 2)2 ( v N / 2) 2 ] 1/ 2 (M/2,N/2): center in frequency domain D0 is called the cutoff frequency. 5 Ideal Low Pass Filtering Low pass filtered image is inverse Fourier Transform of product of F and m Example: Consider the following image and its DFT Image DFT Ideal Low Pass Filtering Applying Image after low pass filter inversion to DFT Cutoff D = 15 low pass filter low pass filter Cutoff D = 5 Cutoff D = 30 Note: Sharp filter Cutoff causes ringing FT ringing and Ideal in frequency blurring domain means non-ideal in spatial domain, vice versa. 8 9 10 Butterworth Lowpass Filters (BLPF) • Smooth transfer function, 1 no sharp discontinuity, H (u,v) 2n D(u,v) no clear cutoff 1 D frequency. 0 1 2 11 Butterworth Lowpass Filters (BLPF) 1 H (u,v) 2n D(u,v) 1 D0 n=1 n=2 n=5 n=20 h(x) 12 No serious ringing artifacts 13 Top: Original image. Bottom: Image filtered with 5th order Butterworth lowpass filter. Gaussian Lowpass Filters (GLPF) • Smooth transfer function, D2 (u,v) smooth impulse 2 response, no ringing H(u,v) e 2D 0 15 GLPF Frequency domain Gaussian lowpass filter Spatial domain 16 No ringing artifacts 17 18 19 Examples of Lowpass Filtering 20 Examples of Lowpass Filtering Low-pass filter H(u,v) Original image and its FT Filtered image and its FT 21 Ideal High Pass Filtering Opposite of low pass filtering: eliminate center (low frequency values), keeping others High pass filtering causes image sharpening If we use circle as cutoff again, size affects results Large cutoff = More information removed DFT of Image after Resulting image high pass Filtering after inverse DFT High-pass Filters • Hhp(u,v)=1-Hlp(u,v) 1 D(u,v) D0 • HIdeal:(u,v) 0 D(u,v) D0 2 1 • Butterworth: | H (u,v) | 2n D0 1 D(u,v) D2 (u,v)/2D2 • Gaussian: H(u,v) 1e 0 23 Ideal High Pass Filtering: Effect of Cutoffs Low cutoff High pass filtering frequency removes of DFT with filter Only few lowest Cutoff D = 5 frequencies High cutoff High pass filtering frequency removes of DFT with filter many frequencies, Cutoff D = 30 leaving only edges Images taken from Gonzalez & Woods, Digital Image Processing (2002) emphasisHigh frequency Original result image Highpass Filtering Example result equalisation filtering Highpass histogram histogram After After 26 High-pass Filters h(x,y) 27 Ideal High-pass Filtering ringing artifacts 28 29 30 Butterworth Filtering Sharp cutoff leads to ringing To avoid ringing, can use circle with more gentle cutoff slope Butterworth filters have more gentle cutoff slopes Ideal low pass filter Butterworth low pass n = 2 n = 4 filter n called order of the filter, controls sharpness of cutoff Butterworth High Pass Filtering Ideal high pass filter Butterworth n = 2 n = 4 hig h pass filter Low Pass Butterworth Filtering Low pass filtering removes high frequencies, blurs image Gentler cutoff eliminates ringing artifact DFT of Image after low Resulting image pass Butterworth filtering after inverse DFT High Pass Butterworth Filtering DFT of Image after high Resulting image pass Butterworth filtering after inverse DFT Butterworth High-pass Filtering 35 • It is possible increase the sharpness of an image by boosting the power in the edges, which means boosting the power in the high frequencies. • This is easily done through a linear combination of the original image with the image resulting from a highpass filter. Top: Original image. Bottom: Image filtered with 3rd order Butterworth sharpening filter, normalized frequency .5, Gp = 1.1, Gs = .8 Gaussian High-pass Filtering Original image Gaussian filter H(u,v) Filtered image and its FT 38 Gaussian High-pass Filtering 39 40 41 42 43 5.4 Periodic Noise Reduction by Frequency Domain Filtering • Periodic noise is due to the electrical or electromechanical interference during image acquisition. • Can be estimated through the inspection of the Fourier spectrum of the image. 5.4 Periodic Noise Reduction by Frequency Domain Filtering • Bandreject filters – Remove or attenuate a band of frequencies. 1 if D(u,v) D0 W / 2 H (u,v) 0 if D0 W / 2 D(u,v) D0 W / 2 1 if D(u,v) D0 W / 2 • D0 is the radius. • D(u, v) is the distance from the origin, and • W is the width of the frequency band. Bandpass filter • Obtained form bandreject filter Hbp(u,v)=1-Hbr(u,v) • The goal of the bandpass filter is to isolate the noise pattern from the original image, which can help simplify the analysis of noise, reasonably independent of image content. Bandreject Filters Periodic Noise Reduction by Frequency Domain Filtering 2D Fourier Transform Examples: Scaling Stretching image => Spectrum contracts And vice versa 2D Fourier Transform Examples: Periodic Patterns Repetitive periodic patterns appear as distinct peaks at corresponding positions in spectrum Enlarging image ( c) causes Spectrum to contract (f) 2D Fourier Transform Examples: Oriented, elongated Structures Man‐made elongated regular patterns in image => appear dominant in spectrum 2D Fourier Transform Examples: Natural Images Repetitions in natural scenes => less dominant than man‐ made ones, less obvious in spectra 2D Fourier Transform Examples: Natural Images Natural scenes with repetitive patterns but no dominant orientation => do not stand out in spectra 2D Fourier Transform Examples: Printed Patterns Regular diagonal patterns caused by printing => Clearly visible/removable in frequency spectrum. Let’s Look at Some Real Images! In this image you have a bunch of cells that are all the This image for example looks ordered but I couldn’t same size but there is no order to their arrangement. tell you exactly what that order is. There are enough of them that they are pretty tightly packed in some regions. After taking a FT of the image it is very apparent This is reflected in the FT image because there is a what sort of order it has and one can determine all circle which represents the average distance they the distances between nearest neighbors just by are from each other but it also shows that there is taking the reciprocal of the distances between a no preferred long range order. dot and the center of the image. The power of FT is that it allows you to take a seemingly complicated image which has an apparent order that is difficult to determine see and break it up into its component sine waves. Fourier Transform Images are from: http://www.cs.unm.edu/~brayer/vision/fourier.html 5.4.3 Notch filters • Notch filter rejects (passes) frequencies in predefined neighborhoods about a center frequency. 0 if D1(u,v) D0 or D2 (u,v) D0 H(u,v) 1 otherwise where 2 2 1/ 2 D1(u,v) (u M / 2u0 ) (v N / 2v0 ) 2 2 1/ 2 D2 (u,v) (u M / 2u0 ) (v N / 2 v0 ) 5.4.3 Notch filters • Butterworth notch filter 1 H (u,v) 2 n D0 1 D1(u,v)D2 (u,v) • Gaussian notch filter 1 D (u,v)D (u,v) 1 2 2 2 H (u,v) 1 e D0 • Note that these notch filters will become highpass when u0=v0=0 Notch filters Example 5.8 Use 1-D Notch pass filter to find the horizontal ripple noise Laplacian in Frequency Domain 2 f (x, y) 2 f (x, y) [ ] (u2 v2)F (u,v) x2 y2 2 2 H1(u,v) (u v ) Frequency Spatial domain 2 2 domain 2 f f f 2 2 Laplacian operator x y 60 61 Subtract Laplacian from the Original Image to Enhance It enhanced Original Laplacian image image output Spatial g(x, y) f (x, y) 2 f (x, y) domain Frequency 2 2 G(u,v) F (u,v) (u v )F (u,v) domain new operator 2 2 H 2(u,v) 1 (u v ) 1 H 1(u,v) Laplacian 62 f 2 f f 2 f 63 64 Unsharp Masking, High-boost Filtering • Unsharp masking: fhp(x,y)=f(x,y)- flp(x,y) One more • H (u,v)=1-H (u,v) parameter to adjust hp lp the enhancement • High-boost filtering: fhb(x,y)=Af(x,y)-flp(x,y) • fhb(x,y)=(A-1)f(x,y)+fhp(x,y) 65• Hhb(u,v)=(A-1)+Hhp(u,v) 66 68 • Wiener filtering incorporates both the degradation function and statistical characteristics of noise into the restoration process • Images and noise are considered as random variables • The objective is to find an estimate ƒ such that the mean square error between them is minimized. • It is assumed that the noise and image are uncorrelated Wiener filtering • In signal processing, the Wiener filter is a filter proposed by Norbert Wiener during the 1940s and published in 1949. • Its purpose is to reduce the amount of noise present in a signal by comparison with an estimation of the desired noiseless signal. • The discrete-time equivalent of Wiener's work was derived independently by Kolmogorov and published in 1941.
Details
-
File Typepdf
-
Upload Time-
-
Content LanguagesEnglish
-
Upload UserAnonymous/Not logged-in
-
File Pages115 Page
-
File Size-