局部对比度归一化层Local Contrast Normalization Layer-N

Halcom

局部对比度归一化层Local Contrast Normalization Layer-N：
可参考链接：http://blog.csdn.net/zouxy09/article/details/10007237

1. function y = localContrastNormalize(I, windowSize)
   
2. % Locally contrast normalizes an image. This is very often used as a
   
3. % pre-processing technique in feature learning. It is described in more
   
4. % detail in, for example,
   
5. %
   
6. % Jarett et al., "What is the Best Multi-Stage Architecture for Object
   
7. % Recognition?" (ICCV 2009)
   
8. %
   
9. % EXAMPLE
   
10. %  G = rgb2gray(imread('data/lenna.jpg'));
    
11. %  y = localContrastNormalize(G);
    
12. %  figure(1); imshow(G,[]); figure(2); imshow(y,[]);
    
13. %
    
14. 
    
15. if(isa(I, 'uint8')), I = double(I) / 255.0; end
    
16. if nargin < 2 || isempty(windowSize), windowSize = 8; end
    
17. 
    
18. nd = ndims(I);
    
19. 
    
20. wSide = normpdf(linspace(-2, 2, windowSize), 0, 1);
    
21. w = wSide'*wSide;
    
22. w = w/sum(sum(w));
    
23. 
    
24. conv2opt= 'same';
    
25. % if removeBorder, conv2opt= 'full'; end
    
26. 
    
27. if nd == 2
    
28.     % grayscale input
    
29.     v = I - conv2(I, w, conv2opt);
    
30.     s = sqrt(conv2(v.^2, w, conv2opt));
    
31.     c = mean(mean(s));
    
32.     y = v./max(c, s);
    
33. else
    
34.     % RGB input
    
35.     v1 = I(:,:,1) - conv2(I(:,:,1), w, conv2opt);
    
36.     v2 = I(:,:,2) - conv2(I(:,:,2), w, conv2opt);
    
37.     v3 = I(:,:,3) - conv2(I(:,:,3), w, conv2opt);
    
38.      
    
39.     s = sqrt(conv2(v1.^2, w, conv2opt) + conv2(v2.^2, w, conv2opt) + conv2(v3.^2, w, conv2opt));
    
40.     c = mean(mean(s));
    
41.     mm= max(c, s);
    
42.     y = cat(3, v1./mm, v2./mm, v3./mm);
    
43. end
    
44. end

复制代码