How to feature extract from an image with black background(without taking the black background in consideration)?

I was scrolling through some answers of the same issue and got the following

C = num2cell(YourImage);
C(~YourImage) = {{}};

This will give you a cell array in which there are empty cells ("nothing") where the background was, and cells containing [1] where the foreground was.And

imagesc(YourImage, 'AlphaData', YourImage)
colormap(gray)
set(gca, 'color', 'none')

to set the Region of interest visible,i did try it myself and didn't get a result,What am i doing wrong and the correct way to use this code.

Answer 1

[Image Analyst]

Looks weird. Very, very weird. I'd say what you're doing wrong is using cells in the first place. I don't see any reason for that. Why put each pixel into a cell? And then make the cells empty where the image is zero? And it's not going to put a 1 in cells with foreground. It's going to have the original gray levels in the cells. I mean, you could simply do the same thing that you say you want (0=background of 0, and 1 = foreground of non-zero) much better and with far, far less memory usage simply by saying

C = (YourImage ~= 0).

Post your image and say what feature in it that you're trying to measure.

[Elias Unk]

All the images shall be similar to those where an RGB lesion is contained somewhere in it the rest is black background . Features i need to extract are mean,standard deviation,variance,skew,kurtosis,entropy for color. And at least the 4 texture ones using graycoprops GLCM.

[Image Analyst]

Just as I thought. No cell arrays needed. Throw away that approach you thought you should use - it's no good. Simply extract each color channel:

% Extract the individual red, green, and blue color channels.
redChannel = rgbImage(:, :, 1);
greenChannel = rgbImage(:, :, 2);
blueChannel = rgbImage(:, :, 3);

Create a mask

mask = redChannel == 0 & greenChannel == 0 & blueChannel == 0;

Then use the mask to get the moments of each color channel:

meanR = mean(redChannel(mask));
stdR = std(redChannel(mask));
varR = var(redChannel(mask));
skewnessR = skewness(redChannel(mask));
kurtosisR = kurtosis(redChannel(mask));

and so on for the other color channels. For the GLCM, simply use graycomatrix with 256 levels and set the first row and first column to zero meaning you want to ignore any pair of pixels where one of them has a value of 0.

[Elias Unk]

I took your approach and after few adjustments like adding double for std,var etc.. and doing imcomplement for the mask i got some results,then used region growing on the initial image and got a more accurate mask with consistent results when i added size to the black background so thanks! And when it comes to GLCM i didn't really understand it as i'm not proficient there,may you explain it more and give an example,Regards.

[Elias Unk]

Knowing that i need to extract the 4 features for 4 angles 0 45 90 135 over three distances 1, 3 and 5 pixel-to-pixel

[Image Analyst]

I should have used ~ to get a mask of the non-black area.

mask = ~(redChannel == 0 & greenChannel == 0 & blueChannel == 0);

Looks like you did dilation. Not sure why you want the mask to grow and include some of the black area.

Attached is a demo of graycomatrix. See if that helps. It's all I have time for right now.

[Elias Unk]

The dilation is actually the result i got using ur code and the first is the one i got using region growing which got me the exact lesion shape,thanks for the demo,i'll read it and post another question if i couldn't sort my issue by the end of the day

[Image Analyst]

There is no dilation in my code. I never told you to call imdilate(). My masking code would give you the exact boundaries like in your top binary image, not the lower one with the square dilation artifacts.

[Elias Unk]

I did try the code again to make sure i didn't make any wrong step and for the following code

redChannel = rgbImage(:, :, 1);
greenChannel = rgbImage(:, :, 2);
blueChannel = rgbImage(:, :, 3);
mask = ~(redChannel == 0 & greenChannel == 0 & blueChannel == 0);

I ended up with the same result,doesn't make sense to me either but try it out yourself.

[Image Analyst]

The two binary images you posted don't look anything like the masked RGB images you posted. The only way my code would look like the larger bottom binary image you posted is if your image is heavily corrupted with JPG artifacts. In that case you may have to check not for exact zero, but some small value, like 9 or something;

mask = ~(redChannel <= 9 & greenChannel <= 9 & blueChannel <= 9);