Number of pixel, image outline and image processing

I'm still new in Matlab. I had segmented gray scale image into four regions like this.

http://i44.tinypic.com/2nq8piw.jpg

I would like to calculate area/no. of pixel for each region within each human image. My idea is drawing outline of each human images and then calculate no. of pixel for white, light gray and dark gray regions in the image. Is there any idea to do that, or better idea? What i learnt and found only the method for binary image, not same with this one.

Thanks in advance.

Answer 1

[Sean de Wolski]

doc bwconncomp
doc regionprops
doc bwarea

[Syahrul Niezam]

Thank you.

I think, those are basic in calculating area etc. But it supposedly for binary image. Am I correct?

[Anton Semechko]

Yep, regionprops is for binary images, but it also accepts labeled images as input, see documentation for bwlabel

[Sean de Wolski]

regionprops will work on a label image (multiple levels of gray where each level is treated as a separate object). My idea for bewconncomp/bwarea is that you could regroup into binary objects - this may not be th ebest approach however and a direct call to regionprops would work.

[Syahrul Niezam]

Thank you for your responds.

Do you mean I should apply bwconncomp on the image, and then regionprops?

[Sean de Wolski]

It depends. Do you want each blob, regardless of the colors in it, to be treated as one object if the pixels are connected? If so, binarize and use bwconncomp and regionprops. If you want the different colors to be grouped together regardless of their location - regionprops directly.

[Syahrul Niezam]

Is it possible to draw outline of each human, and the rest objects, and then calculate area/no. of pixel for white, light gray and dark gray area?

[Syahrul Niezam]

I'm sorry, what I understand is bwconncomp is only for binary image. It's not working above image. Correct me if I'm wrong.

[Walter Roberson]

Sean said "if so, binarize and use bwconncomp". You did not binarize.

binaryimg = logical(YourImage);

[Syahrul Niezam]

I'm sorry for my mistake.

This is my code that I tried:

c = imread('image1_118.jpg');

a=double(c);

b = zeros(size(a));

b = uint8(a<90);

b(a >= 110) = 180;

b(a >= 150) = 200;

b(a >= 200) = 255;

subplot(121);figure(1);imshow(b);

y=logical(b);

C = bwconncomp(y);

CC = labelmatrix(C);

subplot(122);figure(1);imshow(y);

The bwconncomp part does not work well.

[Walter Roberson]

Please be more specific about "does not work well".

You do not appear to be displaying C or CC.

[Syahrul Niezam]

This is the code to binarize the image using logical. The result image is attached too.

http://i41.tinypic.com/33u3mrp.jpg

c = imread('image1_118.jpg');

a=double(c);

b = zeros(size(a));

b = uint8(a<90);

b(a >= 110) = 180;

b(a >= 150) = 200;

b(a >= 200) = 255;

subplot(221);figure(1);imshow(b);

y=logical(b);

%x=uint8(y);

%C = bwconncomp(y);

%CC = labelmatrix(C);

subplot(222);figure(1);imshow(y);

%subplot(223);figure(1);imshow(C);

However, the result of C did not come out.

[Walter Roberson]

Your quantization is wrong. You assign the value 1 where a < 90, the value 0 where a was 90 to 109, the value 180 where a was 110 to 149, the value 200 where a was 150 to 199, and the value 255 where a was 200 to 255.

When you logical() this, the only parts that will come out 0 will be the parts where you had 0 in the matrix -- that is, the places where a was 90 to 109. Everything else in b is non-zero, including where a was < 90.

You need to re-think what value you want for a < 90 and for a from 90 to 109.

[Syahrul Niezam]

Thank you for your responds.

I would like to assign the value 0 (black) for b where a<90, b=180 where 91<a<110; b=200 where 111<a<200; and b=255 (white) where 201<a<255.

If that possible, are logical, bwconncomp and labelmatrix can be applied?

[Walter Roberson]

binmap = uint8([0 180 200 255]);

[counts, binidx] = histc(a(:), [0 90 110 200 256]);

b = reshape( binmap(binidx), size(a) );

And proceed.

Note: you did not indicate what should happen for a = 90 exactly, or a = 110 exactly, or a = 200 exactly or a = 255 exactly. The code I gave here uses 90 <= a < 110, 110 <= a < 200, 200 <= a < 256.

[Syahrul Niezam]

Ops, i did mistake. It supposedly b=0 if a<=90; b=180 when 90<a=<110; b=200 when 110<a<=150; and b=255 when 150<a<=255.

That what's I desired to do so before calculating no. of pixel in same area and so forth.

[Walter Roberson]

Okay in that case use

[counts, binidx] = histc(a(:), [0 91 111 151 256]);

[Syahrul Niezam]

So, the whole code wlii be like this?

binmap = uint8([0 180 200 255]);

[counts, binidx] = histc(a(:), [0 91 111 151 256]);

b = reshape( binmap(binidx), size(a) );

subplot(221);figure(1);imshow(b);

y=logical(b);

x=uint8(y);

C = bwconncomp(y);

CC = labelmatrix(C);

subplot(222);figure(1);imshow(y);

subplot(223);figure(1);imshow(C);

[Syahrul Niezam]

I would like to ask, is there any idea how to do labeling on non-binary image; like image above? What I understand, bwlabel is for binary (black and white) image only. Is it possible to 'recreate' or renovate itself?

[Sean de Wolski]

A label image can be _ANY_ image if you choose to treat it as such. That is, you can treat equal intesities as objects.

Answer 2

[Image Analyst]

See my image segmentation tutorial: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862. I do that exact thing, except for coins instead of humans.

[Syahrul Niezam]

Did you mean the blobsdemo?

[Image Analyst]

Yes.

[Syahrul Niezam]

I had studied through the codes. However, it is for binary image. I am not counting the whole area of human image, but area of each white, light gray and dark gray regions within human image. (I hope my English is understandable).

Is it possible to draw outline of each human, and the rest objects, and then calculate area/no. of pixel for white, light gray and dark gray area?

[Image Analyst]

It is not for a binary image. It uses a gray scale image, just like you do. There is a binary image that defines where you look in the gray scale image to compute areas and intensities, etc. but the original image is still gray scale and you should be able to adapt it very easily to your image. Very easily.

[Syahrul Niezam]

I would like to ask, is there any idea how to do labeling on non-binary image; like image above? What I understand, bwlabel is for binary (black and white) image only. Is it possible to 'recreate' or renovate itself?

[Image Analyst]

You can't do labeling on a non-binary image. It doesn't make sense. You have to have an image that separates your image into foreground and background, and that is a binary image. How can you do anything without deciding what is foreground and background? Tell me, does your image have a foreground? Of course it does.

[Syahrul Niezam]

I want to apply labeling in above image, so that I can calculate no. of pixel for each levels; white, light gray, dark gray and black.

http://i44.tinypic.com/2nq8piw.jpg

However, I would like to calculate separately for each subject(human and non-human).

I really hope any code, basic or idea to do that.

[Image Analyst]

First use bwareaopen() to get rid of small regions. Then say

binaryImage = yourImage >= 100 & yourImage <= 150; % Or whatever it needs for the range you're interested in.

labeledImage = bwlabel(binaryImage);

measurements = regionprops(labeledImage, 'area');