Generating the edge for a higher intensity pixels and not low intensity pixels.

Hello Image Processing Committee ---

I have tried several ways to isolate the high intensity blue pixels fromt he low intensity pixels. using bilateral filtering method and an example shown below. I am unable to create a shape around the image provided below. The shape should be around the higher intensity blue pixels and not the low intensity pixels. Plus, if possible to apply Otsu thresholding in such condition?

Thank you for you help.

Best regards,

% Read the image
close all; clear all;
ImageData = imread('Trial.jpg');

Error using imread>get_full_filename
File "Trial.jpg" does not exist.

Error in imread (line 372)
fullname = get_full_filename(filename);

% Convert the image to grayscale
RGBImage = imgaussfilt(ImageData,3);
GrayImage = RGBImage(:,:,3);

% Convert the grayscale image to double precision
ImageDataGray = im2double(GrayImage);

% Define the threshold values for edge detection
lowThreshold = 0.6;
highThreshold = 0.9;

% Apply Canny edge detection
edgeImage = edge(ImageDataGray, 'Canny', [lowThreshold, highThreshold]);

% Display the images in subplots
figure;

% Original Image
subplot(1, 3, 1);
imshow(ImageData);
title('Original Image');

% Grayscale Image
subplot(1, 3, 2);
imshow(ImageDataGray);
title('Grayscale Image');

% Edge Image
subplot(1, 3, 3);
imshow(edgeImage);
title('Edge Detection');

% Find connected components in the edge image
CC = bwconncomp(edgeImage);
numObjects = CC.NumObjects;

% Process each connected component
for k = 1:numObjects
    % Extract the k-th object
    BW = false(size(edgeImage));
    BW(CC.PixelIdxList{k}) = true;

    % Calculate the properties of the object using regionprops
    P = regionprops(BW, 'Centroid', 'BoundingBox', 'Area', 'Perimeter');

    % Filter out faint intensity objects based on area
    thresholdArea = 50; % Adjust this value as per your requirement
    if P.Area > thresholdArea
        % Extract the necessary properties
        xCenter = P.Centroid(1);
        yCenter = P.Centroid(2);
        Area = P.Area;
        BoundingBox = P.BoundingBox;

        % Plot the boundary and center of the detected object on the image
        figure;
        imshow(ImageData);
        hold on;
        rectangle('Position', BoundingBox, 'EdgeColor', 'r', 'LineWidth', 2);
        plot(xCenter, yCenter, 'g+', 'MarkerSize', 10, 'LineWidth', 2);
        hold off;
    end
end

[ProblemSolver]

@Image Analyst:

As, it can be seen that the low intensity image is in the background, and a high intensity image is in foreground. I need to draw shape for both the intensities separately.

Thank you for your help.

Answer 1

[Image Analyst]

Not sure what you call low intensity. So if the high intensity mask is all pixels above some threshold, then isn't the low intensity mask just the inverse of that? In other words the whole image except the spot in the middle? And the spot in the middle does not have sharp edges so where the edge of it is, is actually a judgment call. You can get different diameters just by specifying different thresholds. I don't think you need bilateral filteringg or edge detection at all. Why do you think you do? Don't spend time doing unnecessary operations. What are you going do to once you have the high and low intensity masks? That will help me decide what operations other than thresholding that you need to do.

[ProblemSolver]

@Image Analyst: Thank you for your response. I apologize for my tardy explanation.

1) For me the blue circle part is the high intensity pixels. For a series of images, these pixels will grow as shown in the right figure. This shows that the speed is different for these high intensity pixels compared to low-intensity pixels. Therefore, if I have the edges for the high-intensity region and edge of the low intensity region, I can derive the relation of speed. between these two regions.

[Image Analyst]

Yeah, so like I thought, you don't need any filtering or edge detection. So you just have two intensity thresholds, one for the low intensity and one for the high intensity. Let's say they are 50 and 200 (or whatever). So then you can just get the area from the histogram,

counts = histcounts(grayImage);
lowIntensityArea = sum(counts(51:200))
highIntensityArea = sum(counts(201:end))

Or you can use bwarea or nnz on the masks

lowMask = (grayImage >= 50) & (grayImage < 200);
highMask = grayImage >= 200;
% Using nnx for a pixel count
lowIntensityArea = nnz(lowMask)
highIntensityArea = nnz(highMask)
% Using bwarea for an area weighted by boundary shape.
lowIntensityArea = bwarea(lowMask)
highIntensityArea = bwarea(highMask)

If you have a lot of isolated low or high intensity pixels inside the zone for high or low (respectively) then you can do things to get rid of those, like bwareafilt (take the largest single blob) or imfill (fill holes in the blob).

[ProblemSolver]

@Image Analyst: Thank you for the response. Based on your suggestion, I wrote the program as below. However, I have to then manually adjust the threshold values, which is not feasible for a set of say 100 images. Is there an alternate way that I can loop it without having to change the thresholding value?

% close all;
% clear;
% Load the image
rgbImage = imread('Cam20872-FromFile_2736.jpg');
%rgbImage = imread('Cam20872-FromFile_2737.jpg'); 

% Convert to grayscale
grayImage = rgbImage(:,:,3);

% Calculate the low and high intensity areas
lowMask = (grayImage >= 0) & (grayImage < 90);
highMask = grayImage >= 235;

% Using nnz for a pixel count
lowIntensityArea_nnz = nnz(lowMask);
highIntensityArea_nnz = nnz(highMask);

% Using bwarea for an area weighted by boundary shape.
lowMask_bw = logical(lowMask);  % Convert to binary image
highMask_bw = logical(highMask);  % Convert to binary image

lowIntensityArea_bwarea = bwarea(lowMask_bw);
highIntensityArea_bwarea = bwarea(highMask_bw);

% Display the results using subplots
figure;

% Original Image
subplot(2, 3, 1);
imshow(grayImage);
title('Gray Image');

% Low Intensity Area using nnz
subplot(2, 3, 2);
imshow(lowMask);
title('Low Intensity Area (nnz)');
xlabel(['Area: ' num2str(lowIntensityArea_nnz)]);

% High Intensity Area using nnz
subplot(2, 3, 3);
imshow(highMask);
title('High Intensity Area (nnz)');
xlabel(['Area: ' num2str(highIntensityArea_nnz)]);

% Low Intensity Area using bwarea
subplot(2, 3, 4);
imshow(lowMask_bw);
title('Low Intensity Area (bwarea)');
xlabel(['Area: ' num2str(lowIntensityArea_bwarea)]);

% Histogram of the grayscale image
subplot(2, 3, 5);
imhist(grayImage);
title('Histogram');

% High Intensity Area using bwarea
subplot(2, 3, 6);
imshow(highMask_bw);
title('High Intensity Area (bwarea)');
xlabel(['Area: ' num2str(highIntensityArea_bwarea)]);

% Adjust the subplot layout
sgtitle('Intensity Area Analysis');

Thank you in advance!

[Image Analyst]

Using a fixed threshold is best. It's the most reliable, robust, and repeatable. If you're not able to use a fixed threshold, you have to ask yourself why. Do you not have good control over your image capture conditions like lighting and camera settings? If not, try to control them so that the picture is largely the same, like the intensity is the same and only the diameter changes. Then you can use a fixed threshold and get an accurate diameter. If you have to change both the threshold because the settings change and also the object itself changes then you will have noisier results because as I already said you can get different diameters depending on the threshold. So the measured diameter would be changing not only due to the subject/object itself changing but also due to adjustments in the threshold. Try to avoid automatic threshold selection if at all possible and correct your image capture situation instead.

[ProblemSolver]

@Image Analyst - Yes, I completely understand. Thats the reason before you wrote this answer, I already accepted your answer. I did check the accuracy of my camera of my camera and the light setting, the skewness, sauration of the pixels, etc. to validate if the camera setting can be refined more. However, the camera setting were set and the parallelisim between the CMOS chip and the window was accurate within an uncertainty was within less than 1.045%.

Thank you for the help.

Answer 2

[DGM]

If you're wanting to identify regions by intensity, why are you using edgefinding? Edgefinding is sensitive to gradient, not to intensity.

inpict = imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/1420743/image.jpeg');

% i'm not sure which areas you're trying to select
% i'm going to select the whitish peak separately from the blue disk
% if the two are to be exclusive sets, then just use boolean ops to combine the two
peakmk = inpict(:,:,1) > 26;
mainmk = inpict(:,:,3) > 103;

imshow(peakmk)

imshow(mainmk)

If you want to reduce those to masks delineating the mask boundary, use bwperim() or bwboundary() on the corresponding mask.

[ProblemSolver]

@DGM Thank you for your response.

But if I chose to perform thresholding the intensity in this way, I would have to manually perform on it every single image file. Is there a way that I don't have to change the hard coded value (26 and 103), but it checks that automatically for each image and then generates the binarized images. If you see this image you have overlapping of bright white light on low intensity white light. But the edge detection is only performed on the outer low-intensity rather than the higher-intensity white light:

peakmk = inpict(:,:,1) > 26;
mainmk = inpict(:,:,3) > 103;

[DGM]

It's hard to come up with a method that will work for varied images based on a single sample image. We'd need to know how the images can plausibly vary -- both the manner in which they vary and the degree to which they vary.

[ProblemSolver]

I understand. The pictures plausible changes are as such shown in the these three images. In the first picture you have only the low intensity pixels, then in the next one you have low and high intensity pixels, and finally you have the low-intensity pixels are almost equivalent to the high-intensity pixels.