LAB image classification using matlab - algorithm

Question

I am trying to implement this algorithm, but I am a little confused about the classification of L, A and B . the algorithm outline is as follows:
Convert the RGB image to a LAB image.
Compute the mean values of the pixels in L, A and B planes of the image
separately.
If mean (A) + mean (B) ≤ 256
3.1. Classify the pixels with a value in L ≤(mean(L) – standard deviation
(L)/3) as shadow pixels and others as non-shadow pixels.
Else classify the pixels with lower values in both L and B planes as shadow
pixels and others as non-shadow pixels.
I am very much confused about how to classify L as a shadow pixel and others as non shadow pixels.
This is the code which has the algorithm implemented:
clear;
I = imread('flower.jpg');
rgb = imresize(I,[256,256]);
mlab = makecform('srgb2lab');
lab = applycform(rgb,mlab);
S_bin = im2bw(rgb);
S = S_bin + (S_bin == 0);
NS = S;
for i=1:255
for j=1:255
A = lab(j,i,2);
B = lab(j,i,3);
L = lab(j,i,1);
Lmean = mean(mean(L));
Amean = mean2(mean2(A));
Bmean = mean2(mean2(B));
if (Amean + Bmean) <= 256
Lstd = std2(L);
std = (Lmean-(Lstd/3));
Lmean = mean(mean(L));
if L <= std
S(j,i,1) = L;
S(j,i,2) = L;
S(j,i,3) = L;
else
S = lab;
end
else
if L < B
S(j,i,1) = L;
S(j,i,2) = L;
S(j,i,3) = L;
else
S(j,i,1) = B;
S(j,i,2) = B;
S(j,i,3) = B;
end
end
end
end
S = lab2rgb(S);
S=uint8(round(S*255));
figure(3);imshow(S);

Answer 1

First of all: L, A and B are scalars. Just one single value. Taking means or standard deviations therefore will not work.
Assign L, A and B first, then take means, standard deviations etc, then built you image in a second loop:
I = imread('flower.jpg');
rgb = imresize(I,[256,256]);
mlab = makecform('srgb2lab');
lab = applycform(rgb,mlab);
S_bin = im2bw(rgb);
S = S_bin + (S_bin == 0);
NS = S;
L = squeeze(lab(:,:,1);
A = squeeze(lab(:,:,2);
B = squeeze(lab(:,:,3);
Lmean = mean(L(:));
Amean = mean(A(:));
Bmean = mean(B(:));
Lstd = std(L(:));
Astd = std(A(:));
Bstd = std(B(:));
if (Amean + Bmean) <= 256
if L<=(Lmean-(Lstd/3));
tmp = L<=(Lmean-(Lstd/3));
S(:,:,1) = zeros(size(L))+L(tmp);
S(:,:,2) = zeros(size(L))+L(tmp);
S(:,:,3) = zeros(size(L))+L(tmp);
else
S = lab;
end
tmp=L<B;
S(:,:,1) = zeros(size(L))+L(tmp);
S(:,:,1) = zeros(size(L))+L(tmp);
S(:,:,1) = zeros(size(L))+L(tmp);
tmp=L=>B;
S(:,:,1) = zeros(size(B))+B(tmp);
S(:,:,1) = zeros(size(B))+B(tmp);
S(:,:,1) = zeros(size(B))+B(tmp);
end
S = lab2rgb(S);
S=uint8(round(S*255));
figure(3);imshow(S);
This way you will just get a one colour pane, either coloured according to L or B, but since you had that before in your code I left it here.