武汉理工大学数字图像处理英文课件 Chapter7（1）.pptVIP

Autocorrelation via FFT Autocorrelation of cyl function Autocorrelation of cyl (1/2) Autocorrelation of cyl (2/2) Autocorrelation Example 2 Autocorrelation vs. Self-Convolution Autocorrelation Example 2 Autocorrelation vs. Self Convolution Matlab Function fftprep Direct Methods * f = auto(a) % f = auto(a) % returns as f the auto-correlation of image a. f = fft2(a); f = f.*conj(f); f = ifft2(f); f = fftshift(f); f = real(f); f = f/max(max(f)); Fourier Transform of gf (x) is |F(x)|2 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 r gcyl(r) function y = gcyl(r) y = zeros(size(r)); idx = find(r1); x = r(idx); y(idx) = (2.0/pi)*(acos(x)-x.*sqrt(1-x.*x)); The marked data points are taken from a horizontal cross-section of the output image. The solid curve is the ideal function, which was generated from ideal(r/8). The factor of 8 converts pixel distance to the appropriate normalized distance. ? x = 0:8; ? y = out(65,x+65); ? plot(x,y); ? xi = linspace(0,8,100); ? yi = ideal(xi/8); ? plot(x,y,o,xi,yi); ? xlabel(radius (pixels)); ? ylabel(value); ? a = imread(single.tif); ? out = auto(a); The agreement between theory and calculation is not as good as we might wish. ? a = imread(circ.tif); ? out = auto(a); ? x = 0:31; ? y = out(33,33+x); ? xi = linspace(0,31,100); ? yi = ideal(xi/32); ? plot(x,y,o,xi,yi); ? xlabel(radius (pixels)); ? ylabel(value); To test the hypothesis that the disagreement is due to sampling error, we increased the size of the circle by four so that its autocorrelation just fits in a 64 x 64 array. The results, shown above, are much more satisfactory. The input function consists of two circles (cylinder functions), one displaced 8 pixels to the left and the other 8 pixels to the right. Each circle is 8 pixels in diameter. A cross-section of the autocorrelation function is shown below There are three peaks. The central peak is twice the height of its neighbors. The relative peak heights and locations are shown in the