How do I find the pixel co-ordinates in an image in order to find the colour properties like red_mean, blue_mean, etc at that pixel?

I want to find the pixel co-ordinates of an image containing nuclei of cells. I implemented the following commands:

conn_c = bwconncomp(tneg_bwLabel);

PixelListTo = regionprops(conn_c,'PixelList');

I received a struct that has a pixel list. How do I find the red mean, blue mean, gray mean and standard deviation at these particular pixel co-ordinates that I have received in the pixel list?

Also, I am getting values like 29 x 1 double in PixelList. Could you help me understand how to find co-ordinates using these values?

Answer 1

[Image Analyst]

You don't need to get the pixel locations of every pixel in the blobs. You just pass in the labeled image itself and ask regionprops for the MeanIntensity. Try this:

% Extract the individual red, green, and blue color channels.
redChannel = rgbImage(:, :, 1);
greenChannel = rgbImage(:, :, 2);
blueChannel = rgbImage(:, :, 3);
% Now find the mean intensities for each blob identified by the labeled image.
propsRed = regionprops(tneg_bwLabel, redChannel, 'MeanIntensity');
propsGreen = regionprops(tneg_bwLabel, greenChannel, 'MeanIntensity');
propsBlue = regionprops(tneg_bwLabel, blueChannel, 'MeanIntensity');

[Rucha Apte]

Thank you so much. I was able to get an putput for red_mean. However the same code isn't working for blue_mean.

[Image Analyst]

Why not? It should. Why do you think the value is wrong?

Attach your image and script.

[Rucha Apte]

I am getting an error saying : Index in position 3 exceeds array bounds (must not exceed 1).

I tried to run the above code on different sections of my code. I am able to get an putput until the stage where i perform nuclei detection. Howver when I given the nuclei detected image to blueChannel = rgbImage(:, :, 3); I am getting an error.

[Image Analyst]

You have a gray scale image. See where it says "must not exceed 1"? That means rgbImage is not an RGB image with separate channels for red, green, and blue but is instead a gray scale image with only one channel. Therefore there will no no red, green, and blue values.

Again, attach your image and script.

[Rucha Apte]

Yes. Thank you so much. It worked. I was giving it a grayscale image as an input. I was now able to find the red mean and blue mean. I have attached my script for Matlab in this comment. Also, will the same commands work by applying the standard deviation command as well?

%FEATURE EXTRACTION
tneg_gray =rgb2gray(negative_nuclei);
tneg_bw = imbinarize(tneg_gray);
tneg_bwLabel = bwlabel(tneg_bw);
cc = bwconncomp(tneg_bwLabel);
s  = regionprops('table',tneg_bwLabel,'all'); % The properties can be included from the regionprops help. Here only two are show e.g. Area, Centroid
% s(s.Area<50,:)=[];
% Extract the individual red, green, and blue color channels.
redChannel = negative_nuclei(:,:,1);
blueChannel = negative_nuclei(:,:,3);
% Now find the mean intensities for each blob identified by the labeled image.
propsRed = regionprops(tneg_bwLabel, redChannel, 'MeanIntensity');
propsBlue = regionprops(tneg_bwLabel, blueChannel, 'MeanIntensity');
 

[Image Analyst]

That doesn't make sense. If it's a gray scale image, you won't have a blue channel. And if it has a blue channel, then it's an RGB image and would also have a green channel. You're using the mask on the red and blue channels, which is fine, but it has to have a green channel also.

Why won't you attach your image? Is it secret or you don't know how (use frame or paper clip icon) or something?

[Rucha Apte]

I gave the RGB image as my input image. I was making the mistake of extracting channels from the grayscale image. Yes, the image has the green channel, however we aren't considering it for feature extraction.Actually, the data is medical data, so I had to take permission from my head while sharing the image. I have attached the image in this comment. Sorry for the delay.

[Image Analyst]

So everything is working now then? It sounds like it. If not, say what help you need.

[Rucha Apte]

Actually I had few doubts.

1. Will I have to use Pixel co-ordinates to find the standard deviation of each blob?
2. Is it possible to horizontally concatenate two tables having different number of rows?

Thank you once again,

[Image Analyst]

1. SD of what? The areas of all the blobs, or the sd of the intensities within each blob?
2. I think you can use join() or vertcat() or semicolon to stitch tables together. Like

tBoth = [t1;t2];  % untested though!

[Rucha Apte]

1. SD of intensities within each blob.
2. join() , vertcat() and horzcat() aren't working as it requires both table t1 and t2 to have same number of rows.