Discover MakerZone

MATLAB and Simulink resources for Arduino, LEGO, and Raspberry Pi

Learn more

Discover what MATLAB® can do for your career.

Opportunities for recent engineering grads.

Apply Today

segmenting an image into number of equal size images

Asked by karuna on 1 Apr 2012

can anybody give me an algorithm for segmenting an image into certain number of equal size image and also for calculating properties for each segmented images?

3 Comments

Image Analyst on 1 Apr 2012

We should really add this to the FAQ. This "how do I divide an image into blocks/subimages" question is getting almost as common as the "how do I save a figure" questions.

M@lik Ali on 16 Jul 2012

yes i think must add this to FAQ.

Jeff on 9 Aug 2013

Check the function bestblk([M N],maxsiz). It finds dimensions <= maxsiz that are divisors of M & N. If it can't find a divisor > M/10, then it minimizes the amount of padding (and then does the same for N).

http://www.mathworks.com/help/images/ref/bestblk.html

karuna

Products

No products are associated with this question.

2 Answers

Answer by Image Analyst on 1 Apr 2012
Accepted answer

See my demo for two different ways to use blockproc to do it:

function blockproc_demo()
try
	clc; % Clear the command window.
	close all; % Close all figures (except those of imtool.)
	workspace; % Make sure the workspace panel is showing.
	fontSize = 20;
	% Change the current folder to the folder of this m-file.
	if(~isdeployed)
		cd(fileparts(which(mfilename)));
	end
	% Read in standard MATLAB demo image.
	grayImage = imread('cameraman.tif');
	[rows columns numberOfColorChannels] = size(grayImage);
	% Display the original image.
	subplot(2, 2, 1);
	imshow(grayImage, []);
	caption = sprintf('Original Image\n%d by %d pixels', ...
	rows, columns);
	title(caption, 'FontSize', fontSize);
	% Enlarge figure to full screen.
	set(gcf, 'Position', get(0,'Screensize'));
	set(gcf, 'name','Demo by ImageAnalyst', 'numbertitle','off')
	% Block process the image.
	windowSize = 3;
	% Each 3x3 block will get replaced by one value.
	% Output image will be smaller by a factor of windowSize.
	myFilterHandle = @myFilter;
	blockyImage = blockproc(grayImage,[windowSize windowSize], myFilterHandle);
	[rowsP columnsP numberOfColorChannelsP] = size(blockyImage);
	% Display the processed image.
	% It is smaller, but the display routine imshow() replicates
	% the image so that it looks bigger than it really is.
	subplot(2, 2, 2);
	imshow(blockyImage, []);
	caption = sprintf('Image Processed in %d by %d Blocks\n%d by %d pixels\nCustom Box Filter', ...
	windowSize, windowSize, rowsP, columnsP);
	title(caption, 'FontSize', fontSize);
	% Now let's do it an alternate way where we use an anonymous function.
	% We'll take the standard deviation in the blocks.
	windowSize = 8;
	myFilterHandle2 = @(block_struct) ...
	std2(block_struct.data) * ones(size(block_struct.data));
	blockyImageSD = blockproc(grayImage, [windowSize windowSize], myFilterHandle2);
	[rowsSD columnsSD numberOfColorChannelsSD] = size(blockyImageSD);
	subplot(2, 2, 4);
	imshow(blockyImageSD, []);
	caption = sprintf('Image Processed in %d by %d Blocks\n%d by %d pixels\nAnonymous Standard Deviation Filter', ...
	windowSize, windowSize, rowsSD, columnsSD);
	title(caption, 'FontSize', fontSize);
	% Note: the image size of blockyImageSD is 256x256, NOT smaller.
	% That's because we're returning an 8x8 array instead of a scalar.
	uiwait(msgbox('Done with demo'));
	catch ME
		errorMessage = sprintf('Error in blockproc_demo():\n\nError Message:\n%s', ME.message);
		uiwait(warndlg(errorMessage));
	end
	return;
% Takes one 3x3 block of image data and multiplies it
% element by element by the kernel and
% returns a single value.
function singleValue = myFilter(blockStruct)
	try
		% Assign default value.
		% Will be used near sides of image (due to boundary effects),
		% or in the case of errors, etc.
		singleValue = 0;
		% Create a 2D filter.
		kernel = [0 0.2 0; 0.2 0.2 0.2; 0 0.2 0];
		% kernel = ones(blockStruct.blockSize); % Box filter.
		% Make sure filter size matches image block size.
		if any(blockStruct.blockSize ~= size(kernel))
			% If any of the dimensions don't match.
			% You'll get here near the edges,
			% if the image is not a multiple of the block size.
			% warndlg('block size does not match kernel size');
			return;
		end
		% Size matches if we get here, so we're okay.
		% Extract our block out of the structure.
		array3x3 = blockStruct.data;
		% Do the filtering. Multiply by kernel and sum.
		singleValue = sum(sum(double(array3x3) .* kernel));
	catch ME
		% Some kind of problem...
		errorMessage = sprintf('Error in myFilter():\n\nError Message:\n%s', ME.message);
		% uiwait(warndlg(errorMessage));
		fprintf(1, '%s\n', errorMessage);
	end
	return; 

0 Comments

Image Analyst
Answer by Aaditya Kalsi on 1 Apr 2012

One way to do this is to decide on a metric to segment the image into. Also, the method I am writing is general and the resulting segments will not completely reconstruct the image, rather extract out N important segments.

Easy Algorithm:

1. Assume a 5x5 (you may choose a different size) block. Select this iteratively and move through the entire image. If you have the Image Processing Toolbox, you may do this by BLOCKPROC.

2. Calculate Variance for each block. You may do this by using VAR() on the block's elements listed out like a vector. Assuming your block is B, then B(:) will list out its elements as a vector.

3. Keep the blocks with the M highest variance values.

Variance is a good starting guess for interesting regions. This however highly depends on your application.

0 Comments

Aaditya Kalsi

Contact us