how to segment hand from an image with non uniform lighting?

I'm trying to segment the hand from the image but due to the non uniform lighting the result come out like this

the result in uniform lighting

how to solve this problem?

Answer 1

[Image Analyst]

Just fix the lighting - it's so simple! It's almost always easier to prevent the problem than to fix it with software and that is definitely the case here.

If you don't want to get the job done in the most sensible, easy, straightforward, and best way (by fixing the illumination), then there are more complicated ways that you can do it, but you have to make some assumptions, like on the size of the hand, or its color, or on the smoothness of the background, or things like that. Some fix you make for one situation might not work for another. Like if you have that picture and then another where the hand is in front of a flesh colored screen, or a screen with a paisley background - those may all take different algorithms.

For what it's worth, I attach a background correction demo.

[H]

Thanks a lot for your answer. Actually the images are from a database called - Cambridge hand gesture data set- the database have 5 sets of images each with different illumination. I'm using this database for a gesture recognition project. by using this code I got a good estimation for the background but the correction done didn't fix the image. Can I use this background model in another way to fix the image?

[Image Analyst]

It sounds like they're intentionally providing poor images for you to test the robustness of your algorithm. Do they provide an algorithm that works for all their images? If so, use that one. If not you'll have to invent your own algorithm, because we don't do complex algorithm development here, just help with syntax, error messages, program flow, etc. If you don't want to invent your own algorithm, then I suggest you use, or start with, one of the algorithms given here: http://www.visionbib.com/bibliography/contentspeople.html#Face%20Recognition,%20Detection,%20Tracking,%20Gesture%20Recognition,%20Fingerprints,%20Biometrics

[H]

Thanks for your help

[Image Analyst]

If you know your background is always neutral colored and your hands are not, then you can convert to hsv color space and threshold for reddish pixels. See my demo : http://www.mathworks.com/matlabcentral/fileexchange/28512-simple-color-detection-by-hue

Answer 2

[Laila Kazemi]

You need adaptive threshold since the background changes. I have not yet found a good code in matlab for this. you might have to find your own algorithm and code it.

[H]

Here I used graythresh(). but I also tried many different levels for the shareholding other than this level but it didn't work. I also used diifferent segmentation techniques and the result is the same. I think the first step should be fixing the illumination.

Answer 3

[Prasad Kalane]

% Do some image illumination correction.

background = imopen(I,strel('disk',15));

I = I - backgroundbackground

I = imadjust(I);

% level = graythresh(I);

bw = im2bw(I,0.3*level);

Answer 4

[Milion Negewo]

background = imopen(I,strel('disk',15));

I = I - backgroundbackground

I = imadjust(I);

% level = graythresh(I);

bw = im2bw(I,0.3*level);

[Image Analyst]

Here is your algorithm:

I = imread('hands.png');
% Display original image.
subplot(2, 3, 1);
imshow(I);
title('Original Image');

% Convert to gray scale so we can call imopen().
I = rgb2gray(I);
% Display original image.
subplot(2, 3, 2);
imshow(I);
title('Gray Scale Image');

background = imopen(I, strel('disk',15));
% Display image.
subplot(2, 3, 3);
imshow(background);
title('Background Image');

I = mat2gray(double(I) - double(background));
I = imadjust(I);
% Display image.
subplot(2, 3, 4);
imshow(I);
title('Background Subtracted Image');

level = graythresh(I);
bw = im2bw(I, 0.3*level);
% Display image.
subplot(2, 3, 5:6);
imshow(bw);
title('Final Binary Image');

The final image doesn't look so good. I think you need to work on it some more.