find circles with different levels of blur

I have an image with circles of different blur levels. I need to count the circles. What is the best way to remove the blurring and find the distinct circles(even the blurred ones). I have tried something but I need to know if theres a better way to do it.

Thanks in advance.

There are both dark circles and bright circles. Heres the image(the original is a tiff but I have converted it to jpeg for viewing)

Heres the jpeg:

As of now I have tried CLAHE:

clahe_img6 = adapthisteq(gray_img,'ClipLimit',0.06,'Distribution','rayleigh');

resultant image:

followed by weiner filters

smoothedImg = wiener2(img,[5 5]);

The resultant image is something like this:

Answer 1

[Garmit Pant]

Hello Ratul Sabui

From what I gather, you're tackling the problem of identifying and counting circles in an image that vary in clarity due to different levels of blurring. Your approach using CLAHE (Contrast Limited Adaptive Histogram Equalization) followed by a Wiener filter is a solid strategy, as it aims to enhance the contrast and reduce the noise, respectively. Further image processing steps should be employed in addition to the ones that you have already used to enhance the quality of detection of the circles.

The following image processing techniques should be applied to the image prior to detecting the circles:

• Step 1: Preprocessing: Continue using CLAHE “adapthisteq” to improve the contrast of the image, which is crucial for bringing out the edges of the blurred circles. However, consider experimenting with different“ClipLimit”values to see if a lower or higher limit yields better results for your specific image.
• Step 2: Noise Reduction: Instead of solely relying on the Wiener filter “wiener2”, you can apply a bilateral filter using “imbilatfilt” before or after the Wiener filter. The bilateral filter is effective at noise reduction while preserving edges, which might help in retaining the outlines of even the most blurred circles.
• Step 3: Edge Enhancement: After noise reduction, applying an edge enhancement technique, such as using the unsharp mask “imsharpen” or a high-pass filter using “imfilter”, can further delineate the circles from the background, making them more detectable by circle detection algorithms.

After applying the above image processing techniques, the circles can be detected using the Hough Circle Transform for detecting circles using the MATLAB function “imfindcircles”. Given the variations in blur levels, you should try adjusting the sensitivity and edge threshold parameters to capture both clear and blurred circles effectively.

The following code snippet applies all the above steps and also iterates over some sample values of the parameters of the functions listed above to demonstrate the variations in detections based on the values of the parameters.

original_img = imread('cropped_image_grey_jpg.jpg'); 
gray_img = im2gray(original_img); 
% Define parameter ranges for experimentation 
wienerSizes = [3, 5]; % Window sizes for Wiener filter 
bilatSpatialSigmas = [1, 2]; % Spatial-domain standard deviation for Bilateral filter 
unsharpAmounts = [0.5, 1]; % Strength of sharpening for Unsharp Mask 
clipLimits = [0.01, 0.03]; % ClipLimit values for CLAHE 
sensitivities = [0.9, 0.95]; % Sensitivity values for circle detection 

% Prepare for grid visualization 
totalCombinations = length(wienerSizes) * length(bilatSpatialSigmas) * length(unsharpAmounts) * length(clipLimits) * length(sensitivities); 
subplotRows = ceil(sqrt(totalCombinations)); 
subplotCols = ceil(totalCombinations / subplotRows); 
figure; 

combinationCounter = 0; 

for wienerSize = wienerSizes 
    for bilatSpatialSigma = bilatSpatialSigmas 
        for unsharpAmount = unsharpAmounts 
            for clipLimit = clipLimits 
                for sensitivity = sensitivities 
                    combinationCounter = combinationCounter + 1; 
                    
                    % Step 1: Apply CLAHE with current ClipLimit 
                    clahe_img = adapthisteq(gray_img, 'ClipLimit', clipLimit, 'Distribution', 'rayleigh'); 
                    
                    % Step 2: Apply Wiener Filter with current size 
                    smoothedImg = wiener2(clahe_img, [wienerSize wienerSize]); 
                    
                    % Step 3: Apply Bilateral Filter with current spatial sigma 
                    smoothedImg = imbilatfilt(smoothedImg, 'DegreeOfSmoothing', bilatSpatialSigma^2 * 2); 
                    
                    % Step 4: Apply Unsharp Mask with current amount 
                    sharpenedImg = imsharpen(smoothedImg, 'Radius', 2, 'Amount', unsharpAmount); 
                    
                    % Detect bright circles with current Sensitivity 
                    [centersBright, radiiBright] = imfindcircles(sharpenedImg, [8 100], 'ObjectPolarity', 'bright', 'Sensitivity', sensitivity, 'EdgeThreshold', 0.1); 
                    numBrightCircles = size(centersBright, 1); % Count of bright circles 
                    
                    % Detect dark circles with current Sensitivity 
                    [centersDark, radiiDark] = imfindcircles(sharpenedImg, [8 100], 'ObjectPolarity', 'dark', 'Sensitivity', sensitivity, 'EdgeThreshold', 0.1); 
                    numDarkCircles = size(centersDark, 1); % Count of dark circles 
                    
                    % Visualize 
                    subplot(subplotRows, subplotCols, combinationCounter); 
                    imshow(sharpenedImg); 
                    hold on; % Keep the image, overlay circles 
                    viscircles(centersBright, radiiBright, 'EdgeColor', 'b'); 
                    viscircles(centersDark, radiiDark, 'EdgeColor', 'r'); 
                    hold off; % Release the hold to ensure next plots are separate 
                    
                    % Title with counts and parameters 
                    titleStr = sprintf('Wiener: %d, Bilat: %.1f, Unsharp: %.1f, Clip: %.2f, Sens: %.2f\nBright: %d, Dark: %d', ... 
                        wienerSize, bilatSpatialSigma, unsharpAmount, clipLimit, sensitivity, ... 
                        numBrightCircles, numDarkCircles); 
                    title(titleStr, 'Interpreter', 'none', 'FontSize', 8); 
                end 
            end 
        end 
    end 
end 

% Adjust subplot spacing 
set(gcf, 'Units', 'Normalized', 'OuterPosition', [0, 0, 1, 1]); % Maximize figure to fit all subplots 

Following is a subset of the outputs generated by the code snippet:

I would suggest you refer to the links to the MathWorks documentation given below:

1. “imfindcircles” function- https://www.mathworks.com/help/releases/R2022b/images/ref/imfindcircles.html
2. “imbilatfilt” function – https://www.mathworks.com/help/releases/R2022b/images/ref/imbilatfilt.html
3. “imsharpen” function: https://www.mathworks.com/help/releases/R2022b/images/ref/imsharpen.html

Hope you find the above explanation and suggestions useful!

[Ratul Sabui]

Thanks for the suggestions. Will definitely try out sharpening.