how to change pixel label based on it's neighbors?

Hello everyone,

I have a label mask contains of (1,2,... Number of labels).

for example:

11112222233333

11122221222233

11122222222233

in some regions I have a wrong labeled pixel for some reason, for instance in region labeled by 2 I have one pixel inside this region labeled by 1. How can I clean these few wrong labeled pixels and give it the value of it's 3x3 neighbours?

Median filter of erosion actually just reorder the kernel but not removing the wrong labelled pixel

I will appreciate any help,

best regards,

[Ameer Hamza]

Can you attach a sample labeled matrix?

[Image Analyst]

Maybe try the new mode filter, modefilt(). If you want to remove "islands" or "holes" in a sea of a different label, then you can do that if you specify the min acceptable hole size and use bwareaopen().

[MatlabUser]

Thanks for replying,I edited the question is it clearer now?

[MatlabUser]

Thanks (Image Analyst) for sharing modefilt(), I'll try it .

Answer 1

[Image Analyst]

Try this demo. It lets you specify the min allowable hole size, so that you cen keep huge holes, like a donut, but remove small holes like salt and pepper noise:

clc;    % Clear the command window.
clear all;
close all;
workspace;  % Make sure the workspace panel is showing.
format short g;
format compact;
fontSize = 14;
fprintf('Beginning to run %s.m ...\n', mfilename);

labeledImage = [
	1,1,1,1,2,2,2,2,2,3,3,3,3,3,
	1,1,1,2,2,2,2,1,2,2,2,2,3,3,
	1,1,1,2,2,2,2,2,2,2,2,2,3,3,
	1,1,1,2,2,2,2,1,2,2,2,2,3,3,
	1,1,1,2,2,2,2,1,2,2,2,2,3,3,
	1,1,1,2,2,2,2,2,2,2,2,2,3,3]
subplot(3, 2, 5);
imshow(labeledImage, [], 'Colormap', lines(4), 'InitialMagnification', 4800);
title('Original labeled Image', 'FontSize', fontSize);
colorbar
axis('on', 'image');
drawnow;
% Define the min acceptable hole size
% Meaning get rid of holes 1 or smaller,
% and keep holes of size 2 or bigger.
minHoleArea = 2;
maxLabel = max(labeledImage(:));
for k = 1 : maxLabel
	% Get a binary image of only this label.
	thisLabel = ismember(labeledImage, k);
	subplot(3, 2, 1);
	imshow(thisLabel, [], 'InitialMagnification', 4800);
	caption = sprintf('Label #%d', k);
	title(caption, 'FontSize', fontSize);
	% Maximize figure window
	if k == 1
		g = gcf;
		g.WindowState = 'maximized';
	end
	
	% Invert it so holes become blobs, and and it with a hole-filled thisLabel
	% so that we get holes only within this particular label.
	filledLabel = imfill(thisLabel, 'holes');
	holeMask = ~thisLabel & filledLabel;
	subplot(3, 2, 2);
	imshow(holeMask, [], 'InitialMagnification', 4800);
	caption = sprintf('All Initial Holes in Label #%d', k);
	title(caption, 'FontSize', fontSize);
	
	% Now size filter these holes to remove holes smaller
	% than the min allowable area.
	holeMask = bwareafilt(holeMask, [1, minHoleArea-1]);
	subplot(3, 2, 3);
	imshow(holeMask, [], 'InitialMagnification', 4800);
	caption = sprintf('Holes smaller than %d\nthat will be deleted from Label #%d', minHoleArea, k);
	title(caption, 'FontSize', fontSize);
	
	% use holeMask to fill in the label
	labeledImage(holeMask) = k;
	subplot(3, 2, 6);
	imshow(labeledImage, [], 'Colormap', lines(4), 'InitialMagnification', 4800);
	caption = sprintf('Labeled Image with\nholes smaller than %d\nnow deleted from Label #%d', minHoleArea, k);
	title(caption, 'FontSize', fontSize);
	
	drawnow;
	promptMessage = sprintf('Do you want to Continue processing,\nor Quit processing?');
	titleBarCaption = 'Continue?';
	buttonText = questdlg(promptMessage, titleBarCaption, 'Continue', 'Quit', 'Continue');
	if contains(buttonText, 'Quit', 'IgnoreCase', true)
		break;
	end
end

fprintf('Done running %s.m ...\n', mfilename);

[MatlabUser]

Thank you so much for your replying and explaining. I accepted this answer.

I also would like to know if modefilt() is alternative for this solution because I used it for labelled mask and I notice it gives much smoother output, I mean it did what I wanted also to remove wrong labelled pixels that are different than their neighbors.

[Image Analyst]

You can try modefilt(). It probably will give a smoother image, if you think that a smoother image is more accurate. It will not give you precise control over the size of the holes you want to keep or exclude like my code will.