Image Segmentation Using Split and Merge

I have a grayscale image that I would like to segment.

I would like to write a script and predicate that uses the split and merge method for this segmentation.

I would also like to be able to control the minimum block size.

Answer 1

[Akshat]

Hi @Victor Fletcher,

As per the query, I think you just want to implement the split and merge method for the segmentation.

Please refer to the following code to get a rough idea of how I am thinking of implementing the recursive split.

function segmentedImg = splitAndMerge(img, minBlockSize, threshold)
    img = im2double(img);
    
    % Initialize the segmented image
    segmentedImg = zeros(size(img));
    
    % Start the recursive split and merge process
    segmentedImg = recursiveSplit(img, segmentedImg, 1, 1, size(img, 1), size(img, 2), minBlockSize, threshold);
end

function segmentedImg = recursiveSplit(img, segmentedImg, x, y, width, height, minBlockSize, threshold)
    % Extract the current block
    block = img(x:x+width-1, y:y+height-1);
    
    % Calculate the block's mean and standard deviation
    blockMean = mean(block(:));
    blockStd = std(block(:));
    
    % If the block is uniform enough or smaller than the minimum size, fill it
    if blockStd < threshold || width <= minBlockSize || height <= minBlockSize
        segmentedImg(x:x+width-1, y:y+height-1) = blockMean;
        return;
    end
    
    % Otherwise, split the block into four quadrants and process each recursively
    halfWidth = floor(width / 2);
    halfHeight = floor(height / 2);
    
    % Top-left quadrant
    segmentedImg = recursiveSplit(img, segmentedImg, x, y, halfWidth, halfHeight, minBlockSize, threshold);
    % Top-right quadrant
    segmentedImg = recursiveSplit(img, segmentedImg, x, y + halfHeight, halfWidth, height - halfHeight, minBlockSize, threshold);
    % Bottom-left quadrant
    segmentedImg = recursiveSplit(img, segmentedImg, x + halfWidth, y, width - halfWidth, halfHeight, minBlockSize, threshold);
    % Bottom-right quadrant
    segmentedImg = recursiveSplit(img, segmentedImg, x + halfWidth, y + halfHeight, width - halfWidth, height - halfHeight, minBlockSize, threshold);
end

An example driver code for the above functions can be:

img = imread('your_image.png');

minBlockSize = 8; % Define the minimum block size
threshold = 0.05;
segmentedImg = splitAndMerge(img, minBlockSize, threshold);

imshow(segmentedImg, []);
title('Segmented Image');

I hope the above written functions will be able to help you with the split and merge method of segmentation.

[Image Analyst]

image is a built-in function name, so you should not use it as the name of a variable.

[Akshat]

Thanks for pointing that out @Image Analyst! I have updated the answer.