How to extract eye from a facial image?

Hi. I have a big problem. I wanna extract eyes from a facial image an I've written some codes for it. Unfortunately, the result is not reasonable. for the first step, I need to choose the eye's region. The result of this stage must be at the line 50, means subplot(4,4,10). But it's not acceptable and I don't have the eye region. Would any one please tell me what's the problem here?

if true
  % i=imread('pic8.jpg');
subplot(4,4,1)
imshow(i)
title('original image');
iycbcr=rgb2ycbcr(i);
iycbcr = im2double(iycbcr);
subplot(4,4,2)
imshow(iycbcr)
title('YCBCR space');
y=iycbcr(:,:,1);
cb=iycbcr(:,:,2);
cr=iycbcr(:,:,3);
ccb=cb.^2;
subplot(4,4,3)
imshow(ccb)
title('CB^2');
ccr=(1-cr).^2;
subplot(4,4,4)
imshow(ccr)
title('(1-CR)^2');
cbcr=ccb./cr;
subplot(4,4,5)
imshow(cbcr)
title('CB/CR');
igray=rgb2gray(i);
subplot(4,4,6)
imshow(igray)
title('Gray space');
g=1./3;
l=g*ccb;
m=g*ccr;
n=g*cbcr;
eyemapchr=l+m+n;
subplot(4,4,7)
imshow(eyemapchr)
title('Chrom Eyemap');
J=histeq(eyemapchr);
subplot(4,4,8)
imshow(J)
title('Equalized image');
SE=strel('disk',15,8);
o=imdilate(igray,SE);
p=1+imerode(igray,SE);
eyemaplum=o./p;
subplot(4,4,9)
imshow(eyemaplum)
title('Lum Eyemap');
%the eye region
cc=and(J,eyemaplum);
subplot(4,4,10)
imshow(cc)
title('AND of Lum&Chrom');
end

[Zahra kamkar]

This is the pic: http://www.mediafire.com/view/zaawemny6yg4bim/pic8.jpg

[Zahra kamkar]

Salam. I changed a part of the codes and thankfully I obtained a reasonable result. I don't know whether it is right or not because it only responds to the image "pic8.jpg" and for other images, its result is not acceptable!

[Zahra kamkar]

to get the image i=imread('pic8.jpg'); subplot(4,4,1) imshow(i) title('original image'); %to transmit the image to YCBCR color space iycbcr=rgb2ycbcr(i); %to have a correct range for y and cb and cr iycbcr = im2double(iycbcr); subplot(4,4,2) imshow(iycbcr) title('YCBCR space'); %to obtain the values of y and cb and cr y=iycbcr(:,:,1); cb=iycbcr(:,:,2); cr=iycbcr(:,:,3); %these are the parts of the formula for chrominance eyemap ccb=cb.^2; subplot(4,4,3) imshow(ccb) title('CB^2'); ccr=(1-cr).^2; subplot(4,4,4) imshow(ccr) title('(1-CR)^2'); cbcr=ccb./cr; subplot(4,4,5) imshow(cbcr) title('CB/CR'); %to transmit the image to gray scale space igray=rgb2gray(i); igray=~igray; subplot(4,4,6) imshow(igray) title('Gray space'); g=1./3; l=g*ccb; m=g*ccr; n=g*cbcr; %% %this is the formula for chrominance eyemap eyemapchr=l+m+n; subplot(4,4,7) imshow(eyemapchr) title('Chrom Eyemap'); %to obtain better result, we'd better get a histogram equalization from chrominance eyemap J=histeq(eyemapchr); subplot(4,4,8) imshow(J) title('Equalized image'); %the parts of formula for obtaining luminance eyemap SE=strel('disk',15,8); o=imdilate(igray,SE); p=1+imerode(igray,SE); eyemaplum=o./p; subplot(4,4,9) imshow(eyemaplum) title('Lum Eyemap'); %as the paper had said, in this section, we must perform AND on the afformentioned eyemaps(lum and chr), now we have the eye approximate refion as a mask cc=and(J,eyemaplum); subplot(4,4,10) imshow(cc) title('AND of Lum&Chrom'); %% subplot(4,4,11) %In this section, we wanna apply the mask (and of lum and chr eyemaps) on the original image to see only the exact eyes imshow(i); i=im2double(i); title('original image'); R=i(:,:,1); G=i(:,:,2); B=i(:,:,3); %because our image is 3D, we should use repmat command to multiply the mask by the original image newi=i.*repmat(cc,[1,1,3]); subplot(4,4,12) imshow(newi) title('Multiplication');

Answer 1

[Image Analyst]

You're making us work blind (pun intended). You forgot to attach pic8.jpg.

Also, it's not so obvious where the boundaries of the eye are. Different people may have different definitions of "eye" and draw outlines in different locations. Do you mean just the "wet" part or are you including the eyelid, eye socket, or eyebrow? Speaking of which, can you just have your user draw the eye region with imfreehand()? See attached demos of imfreehand().

[Image Analyst]

Regarding your comment elsewhere. Correct, because your algorithm is not very robust and you fine tuned it to work on just one particular image. You need to make a more robust algorithm that works for more images, like the people who have published papers on the subject have successfully done. Go here to see their papers http://iris.usc.edu/Vision-Notes/bibliography/people898.html#Finding%20Facial%20Features,%20Eye%20Detection Pick one and code it up and see if it does better than yours. Also read the FAQ on that.

Answer 2

[Zahra kamkar]

Yea, I forgot to send the Image. I wanna detect the eye with eyelids. I used this algorithm: eyemaplum=dilation/(erosion+1). eyemaplchr=(1/3(cb^2+(255-cr)^2+(cb/cr)).eye region(I used cc instead)=AND(eyemaplum,eyemapchr).

[Image Analyst]

You still forgot. Please don't add "Answers" to your question. Add comments to our Answers.

[Zahra kamkar]

This is my pic: http://www.mediafire.com/view/zaawemny6yg4bim/pic8.jpg

Answer 3

[Zahra kamkar]

I sent you the region I need to detect. Thank you in advance

[Image Analyst]

It's still not here.

Answer 4

[Zahra kamkar]

Isn't there anyone to help me? Please!!!

[Image Analyst]

Maybe people are waiting for you to attach the image so they can run your code. Use the image icon or the paperclip icon or both.

[Zahra kamkar]

Hi, This is the pic I used. Unfortunately I can only use my codes on this image. My codes don't work on other images. http://www.mediafire.com/view/zaawemny6yg4bim/pic8.jpg

Answer 5

[Mohsin Raza]

In this project you have to write a MATLAB code that takes the provided image “plain.jpg” as input. You have to implement Bug Eye Algorithm on the input image. Your output should be similar to the sample output image that is given below

[Image Analyst]

How is this an answer to Zahra's 6 year old question? He did not provide a "plain.jpg" image. So why do you think this Bug Eye Algorithm you heard of, and suggested he use, will work with his image?