1 簡介
圖像增強是數字圖像的預處理,對圖像整體或局部特征能有效地改善.我們讨論了基于蝙蝠算法優化直方圖增強的參數;同時,以一個灰階圖像為例,用MATLAB語言實作了蝙蝠算法優化直方圖均衡化和規定化增強處理。
2 部分代碼
%In this code, the two controlling parameters are lamda and gamma,
%increasing lamda increases contrast whereas increasing gamma,
%slowly(usually between 1000-100000) retains the overbrightness of white
%pixels.
%in original paper, the scholar hasn't used lamda in cdf function,
%hence thereby, restricting himself to increase lamda
%line 53 and 54 contains lamda and gamma respectively
% clc
clear all
for p=1:255
for q=1:256
if p==q
D(p,q) = -1;
elseif q==p+1
D(p,q) = 1;
else
D(p,q) = 0;
end
end
end
clc
i = imread('ImgOrig.png');
r = i(:,:,1);
g = i(:,:,2);
b = i(:,:,3);
figure,imshow(i)
title('Original Image')
%for n=1:256
% histogram(1,n) = 0;
%for l = 1:size_of_image(1)
% for b = 1:size_of_image(2)
% if grays(l,b)==n
% histogram(1,n) = histogram(1,n)+1;
% end
%end
%end
%end
%x = 1:1:256;
%figure, plot(x,histogram)
[freqr xr] = imhist(r);
[freqg xg] = imhist(g);
[freqb xb] = imhist(b);
size_of_image = size(r);
number_of_pixels = size_of_image(1)*size_of_image(2);
lamda = 5;%variable to determine the amount of contrast
gamma = 50000 ;
% smoothing_factor = inv(((1+lamda).*eye(256) + gamma.*transpose(D)*D));
% for n = 0:1:255
% nfreqr(n+1,1) = (freqr(n+1,1) + lamda*n);
% nfreqg(n+1,1) = (freqg(n+1,1) + lamda*n);
% nfreqb(n+1,1) = (freqb(n+1,1) + lamda*n);
% end
%
% freqr = smoothing_factor*nfreqr;
% freqg = smoothing_factor*nfreqg;
% freqb = smoothing_factor*nfreqb;
hir = freqr; %for R
[freqr,fminr] = bat_algorithm_image(20,1000, 0.5 ,0.5,hir,lamda,gamma,D,xr);
freqr = transpose(freqr);
hig = freqg; %for G
[freqg,fming] = bat_algorithm_image(20 ,1000, 0.5 ,0.5,hig,lamda,gamma,D,xg);
freqg = transpose(freqg);
hib = freqb; %for B
[freqb,fminb] = bat_algorithm_image(20 ,1000, 0.5 ,0.5,hib,lamda,gamma,D,xb);
freqb = transpose(freqb);
for n = 0:1:255 %probablity Density Function
p(n+1,1) = freqr(n+1,1)/number_of_pixels;
end
cr = zeros(256,1); %cumulative density function
cg = zeros(256,1);
cb = zeros(256,1);
cr(1,1) = freqr(1,1);
cg(1,1) = freqg(1,1);
cb(1,1) = freqb(1,1);
for n = 1:1:255
cr(n+1,1) = cr(n,1) + freqr(n+1,1);
cg(n+1,1) = cg(n,1) + freqg(n+1,1);
cb(n+1,1) = cb(n,1) + freqb(n+1,1);
end
for n = 0:1:255
cr(n+1,1) = cr(n+1,1)/number_of_pixels;
cg(n+1,1) = cg(n+1,1)/number_of_pixels;
cb(n+1,1) = cb(n+1,1)/number_of_pixels;
end
cdfr = (lamda+1)*(255.*cr + 0.5);
cdfg = (lamda+1)*(255.*cg + 0.5);
cdfb = (lamda+1)*(255.*cb + 0.5);
cdfr = round(cdfr);
cdfg = round(cdfg);
cdfb = round(cdfb);
%main Image
main_image = uint8(zeros(size_of_image(1),size_of_image(2),3));
c = 1;
for l = 1:size_of_image(1)
for w = 1:size_of_image(2)
main_image(l,w,c) = cdfr(r(l,w)+1,1);
main_image(l,w,c+1) = cdfg(g(l,w)+1,1);
main_image(l,w,c+2) = cdfb(b(l,w)+1,1);
end
end
figure, imshow(main_image),title('Histogram Modified')
figure, hist(main_image(:,:,1),xr);
hold on
hist(main_image(:,:,2),xg);
hist(main_image(:,:,3),xb);
entropy_of_original_image = entropy(i)
entropy_of_image = entropy(main_image)
mean_Optimized = mean2(main_image);
var_optimzed = std2(main_image);
D = abs(uint8(main_image) - uint8(i)).^2;
mse = sum(D(:))/numel(main_image);
psnr = 10*log10(255*255/mse);
%mae = meanAbsoluteError(main_image,i)
%E = eme(main_image,size_of_image(1),5) 3 仿真結果
