How to split an image into several curved shape small images?

I have an image and I want to divide it into small curved shape image without using manual cropping. For example the attached input will be splited into curved shape small region which are indicated by the line.

Thanks for the help in advance.

[Bjorn Gustavsson]

My code snippet were an example for how you would calculate/extract one such curved segment - provided that the boundary had a shape proportional to cos(x*2*pi/n_hor_pix) and of a known width. For you to use that snippet, you would have to come up with a way to generate the segment boundaries in a neat fashion - I expect that you don't want the lines you have drawn (I guess) in red above and that those are more representative of what you want, If you can model those with some kind of simple sin/bell-type curve you could modify the generation of the y-coordinate in my example reasonably easy. If I've interpreted your question wrong then you might have to explain more precisely what you want.

Answer 1

[Image Analyst]

Try this:

% Demo to find layers in a cross sectional image.
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 20;

%--------------------------------------------------------------------------------------------------------
%    READ IN IMAGE
grayImage = imread('C2_down5um_246.jpg');

% Get the dimensions of the image.
% numberOfColorChannels should be = 1 for a gray scale image, and 3 for an RGB color image.
[rows, columns, numberOfColorChannels] = size(grayImage)
if numberOfColorChannels > 1
	% It's not really gray scale like we expected - it's color.
	% Extract the red channel (so the magenta lines will be white).
	grayImage = grayImage(:, :, 1);
end

%--------------------------------------------------------------------------------------------------------
% Display the image.
subplot(2, 2, 1);
imshow(grayImage, []);
axis('on', 'image');
title('Original Gray Scale Image', 'FontSize', fontSize, 'Interpreter', 'None');
impixelinfo;
hFig = gcf;
hFig.WindowState = 'maximized'; % May not work in earlier versions of MATLAB.
drawnow;

% Binarize
lowThreshold = 35;
highThreshold = 255;
% [lowThreshold, highThreshold] = threshold(lowThreshold, highThreshold, grayImage);

mask = grayImage > lowThreshold;
% Take largest blob
mask = bwareafilt(mask , 1);
% Display the image.
subplot(2, 2, 2);
imshow(mask, []);
axis('on', 'image');
title('Mask', 'FontSize', fontSize, 'Interpreter', 'None');
impixelinfo;

% Scan to get the top row.
[rows, columns] = size(mask);
% Find top curve first.
topRows = nan(1, columns);
for col = 1 : columns
	t = find(mask(:, col), 1, 'first');
	if ~isempty(t)
		topRows(col) = t;
	end
end
hold on;
plot(topRows, 'r-', 'LineWidth', 3)

% Display the image.
subplot(2, 2, 3);
imshow(grayImage, []);
axis('on', 'image');
title('Original Gray Scale Image', 'FontSize', fontSize, 'Interpreter', 'None');
impixelinfo;
drawnow;
hold on;

% Make a bunch of them, each lower than the previous.
deltay = round(rows/8)
allCurves = nan(7, columns);
nonObjectIndexes = isnan(topRows);
for k = 1 : 7
	% Get this curve.
	thisRow = topRows + (k-1) * deltay;
	% Clip it to the bottom of the image.
	thisRow = min(thisRow, rows);
	% Make columns where there was no object NaN again - they were 365.
	thisRow(nonObjectIndexes) = nan;
	allCurves(k, :) = thisRow;
	plot(thisRow, 'r-', 'LineWidth', 3)
end

Not sure what you want to do after that though. And is it really necessary to extend the curves all the way to the left and right edge of the image when there's nothing there? If so, we could fit the curve to a quadratic with polyfit() and extend it that way.

[Dynamic]

@Image Analyst Thanks for your code. However, I did not got the result that I wanted. Basically these red line or curved on the image is not my goal. My probable output are the images between two redline without that line, I have just tried to represent by red line in terms of spliting points. I hope these code must be help to do this.

Thanks a lot

[Bjorn Gustavsson]

@Dynamic Jarjish: You have 2 persons spending time trying to help you, but we have failed so far. This is because you have not precisely enough explained how and by what criteria you make up your red-line-boundaries. Are they supposed to be arbitrarily scribbled curves, different widths and span for each column of each segment for each image? For us, or anyone else to be able to give you the solution you want to this problem, you will have to explain the procedure of creating these boundaries with sufficient precision, then it will be possible to cook up a solution, before that guess-working our way towards a solution will be a very inefficient waste of time.

[Dynamic]

@Bjorn Gustavsson, Thanks for your reply and sorry for misinterpretation. There is no particular creteria to spilt the image in curved shape. if the image between the line are 5*columns then it will be more better for my project. From the input image I want to splite it in several curved image which are laied by the red line. For more information please find the attached for more clarification. My probable output is regions of 1 image, 2 image , and son on. Thanks for your helping hand

[Bjorn Gustavsson]

Then you should be able to work from my template and modify it to at least extract regions 2 to n.

1, simply modify the equation:

y = y + 130  -30*cos(x*2*pi/234); % This is just an illustrating example for your curve

To something that extends over the entire x-range of your image and lies suitably around your first (or second) image region. You will have to change /234 to /sx where sx is size(Im,2). Then also 130 and 30 to better align with your desired boundary. You might even have to switch out cos to some other function for the shape of your region.

2, Then you should be good to go. Give it a try. Show that you've tried something.

[Dynamic]

@Bjorn Gustavsson, your code provides me good result, however, pcolor plot provide me another colormap regions rather than the original grayscale image. So, I have to modify these line also. But I did not understand how to do that?. Although I have seperated several regions by differnt colors, I want the orignal grayscale image for every rgions.

Thanks for your help.

[Bjorn Gustavsson]

OK, then you might only need to change the colormap from the default (parula these days) to gray. Telling matlab to do so is easy:

colormap(gray)

HTH

[Image Analyst]

@Dynamic Jarjish, sorry I wasn't clear what you want (still not sure). I thought if you wanted each band you'd be able to figure it out. So here is how to get each mask and the strip of image in just that mask layer. I'm showing what it looks like after I've extracted the second layer.

% Demo to find layers in a cross sectional image.
clc;    % Clear the command window.
close all;  % Close all figures (except those of imtool.)
clear;  % Erase all existing variables. Or clearvars if you want.
workspace;  % Make sure the workspace panel is showing.
format long g;
format compact;
fontSize = 20;

%--------------------------------------------------------------------------------------------------------
%    READ IN IMAGE
grayImage = imread('C2_down5um_246.jpg');

% Get the dimensions of the image.
% numberOfColorChannels should be = 1 for a gray scale image, and 3 for an RGB color image.
[rows, columns, numberOfColorChannels] = size(grayImage)
if numberOfColorChannels > 1
	% It's not really gray scale like we expected - it's color.
	% Extract the red channel (so the magenta lines will be white).
	grayImage = grayImage(:, :, 1);
end

%--------------------------------------------------------------------------------------------------------
% Display the image.
subplot(2, 3, 1);
imshow(grayImage, []);
axis('on', 'image');
title('Original Gray Scale Image', 'FontSize', fontSize, 'Interpreter', 'None');
impixelinfo;
hFig = gcf;
hFig.WindowState = 'maximized'; % May not work in earlier versions of MATLAB.
drawnow;

% Binarize
lowThreshold = 35;
highThreshold = 255;
% [lowThreshold, highThreshold] = threshold(lowThreshold, highThreshold, grayImage);

mask = grayImage > lowThreshold;
% Take largest blob
mask = bwareafilt(mask , 1);
% Display the image.
subplot(2, 3, 2);
imshow(mask, []);
axis('on', 'image');
title('Mask', 'FontSize', fontSize, 'Interpreter', 'None');
impixelinfo;

% Scan to get the top row.
[rows, columns] = size(mask);
% Find top curve first.
topRows = nan(1, columns);
for col = 1 : columns
	t = find(mask(:, col), 1, 'first');
	if ~isempty(t)
		topRows(col) = t;
	end
end
hold on;
plot(topRows, 'r-', 'LineWidth', 3)

% Display the image.
subplot(2, 3, 3);
imshow(grayImage, []);
axis('on', 'image');
title('Original Gray Scale Image', 'FontSize', fontSize, 'Interpreter', 'None');
impixelinfo;
drawnow;
hold on;

% Make a bunch of them, each lower than the previous.
deltay = round(rows/8)
allCurves = nan(7, columns);
nonObjectIndexes = isnan(topRows);
xFull = 1 : columns;
for k = 1 : 7
	% Get this curve.
	thisRow = topRows + (k-1) * deltay;
	% Clip it to the bottom of the image.
	thisRow = min(thisRow, rows);
	% Make columns where there was no object NaN again - they were 365.
	thisRow(nonObjectIndexes) = nan;
	allCurves(k, :) = thisRow;
	subplot(2, 3, 3);
	plot(thisRow, 'r-', 'LineWidth', 3)
	% Get the band of image data between this and the next row.
	goodIndexes = ~isnan(thisRow);
	x = xFull(goodIndexes);
	y = thisRow(goodIndexes);
	xOutline = [x, fliplr(x)];
	yOutline = [y, fliplr(y + deltay)];
	mask = poly2mask(xOutline, yOutline, rows, columns);
	
	% Display the mask image.
	subplot(2, 3, 4);
	imshow(mask, []);
	axis('on', 'image');
	caption = sprintf('Mask for Band #%d', k);
	title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
	impixelinfo;
	drawnow;
	hold on;
	
	% Mask the image
	maskedImage = grayImage; % Initialize
	maskedImage(~mask) = 0;
	% Display the masked image.
	subplot(2, 3, 5);
	imshow(maskedImage, []);
	axis('on', 'image');
	caption = sprintf('Masked Image in Band #%d', k);
	title(caption, 'FontSize', fontSize, 'Interpreter', 'None');
	impixelinfo;
	drawnow;
	hold on;
	
	promptMessage = sprintf('Showing band %d.\n\nDo you want to Continue processing,\nor Quit processing?', k);
	titleBarCaption = 'Continue?';
	buttonText = questdlg(promptMessage, titleBarCaption, 'Continue', 'Quit', 'Continue');
	if contains(buttonText, 'Quit', 'IgnoreCase', true)
		return; % or break or continue.
	end	
end

But again, I'm not sure what you want to do once you have each masked band, like analyze it, crop it, save it, or something else.

[Dynamic]

@Image Analyst, This great and perfect to my project. Actually, I want to analyse each of the layer and then again reconstruct the whole image. If there is any way, I can reconstruct the image using all maskimage (layers) then my small project will be done. Thanks for your contribution. Now, please can you please suggest me , how can I reconstruct the whole image using these layers image.

Thanks a lot for your code.

[Bjorn Gustavsson]

Set the region outside the masked images to zero, then simply add the different mask-images together.

[Image Analyst]

Like Bjorn said, just add them together. Outside the masked region is already zero so no need to erase them again. Actually no need to reconstruct because you still have the original image, don't you? Well, okay you can erase above the top layer if you want.

To analyze, we need to know what kind of analysis you need to do, like get the mean intensity or something?

meanOfBand = mean(grayImage(mask))

[Dynamic]

@Image Analyst, I want to add some noise for each band and then reconstruct the whole image by using that noisy band.

[Image Analyst]

You can either add noise to the original image, or each sub image by itself. Should be no problem. If my answer does what you want, could you please "Accept this answer"? Thanks in advance.

[Dynamic]

@Image Analyst, Thanks for your support. Yes, I can add noise to whole image, however, I wanted to add different varience of noise for each band. Thanks for your code.

However, still, I have a question to sort out. If I want to 50% overlap band, then how to modify this code? Please, can you modified the code for that?

Thanks a lot

[Image Analyst]

Just have 2 delta y variables. One for moving it down half a band, and the other for the full width of the band. If you really can't figure it out, let me know.

[Dynamic]

@Image Analyst, It will be a great help, if you figure out these. Beacuse this a small small part of the big project. So, I have to complete all the other portions asap. So, please provide me a code that will be a appreciation helpful for me.

Thanks a lot for your helping hand.

[Image Analyst]

I think you could have done it if you had actually tried. But anyway, the solution is attached.

[Dynamic]

@Image Analyst, Thanks a lot.

Answer 2

[Bjorn Gustavsson]

You should be able to do that with interp2, you just have to select the coordinates of your curved regions to use for interpolating points. Something like this:

Im = peaks(234);
y = 1:23;                        
x = 1:234;
[x,y] = meshgrid(x,y);
y = y + 130  -30*cos(x*2*pi/234); % This is just an illustrating example for your curve
subplot(2,2,1)
imagesc(Im)
hold on
plot(x(1,:),y(1,:),'k.')    
plot(x(end,:),y(end,:),'k.')
peaksCurved = interp2(1:234,1:234,Im,x,y);
subplot(2,2,2)
imagesc(peaksCurved)
caxis([-6.5493, 8.106])
subplot(2,2,3)
pcolor(x,y,peaksCurved),caxis([-6.5493, 8.106]),shading flat
axis ij
axis([1 234 1 234])

HTH

[Dynamic]

Thanks for quick response. I think that it will be help me a lot. However, I have two question. 1. The size of my input is not same i.e. (253*253) and 2. The pcolor did not provide me the original grayscale colored curved shape image. Please help me to sort them

[Bjorn Gustavsson]

You will have to replace Im in my code-snippet with your original image. Then change the definition of x and y to match your desired regions, where x ought to span the entire horizontal extent of your image, and y will have to be modified to give you the y-range of your curved region - I made a somewhat naive sin-shaped region where y(x) varied as cos with one period over the size of my simple example-image. The example was intended for you to modify such that it did fit your image and curved shapes (since you did not specify exactly what they were supposed to be I chose not to guess a too detailed shape.). The variable peaksCurved is the intensities of the illustrating-sample-image at the points [x,y] that seemed to be what you asked for.

HTH

[Image Analyst]

@Bjorn Gustavsson, I think he needs help in finding the curve in the first place - that's the main problem.

@Dynamic Jarjish, again, post your original image without the red curves overlaid on it. If we want to write code for you, those red lines really mess it up. How can we find the top surface of your object if we can't see it? And what do you mean by "The pcolor did not provide me the original grayscale slice and small curved shape image."? The pcolor() function, which I don't believe you should use, doesn't return anything. It's you who provide it with the image, and it's that image that I want you to attach. And do you have two images? I'm confused when you said you thought pcolor returned both a "original grayscale slice" image, and "small curved shape image". So are there two images?

[Bjorn Gustavsson]

@Image Analyst, then out of all curves we can draw we'll need to know by what criteria we should chose...

[Dynamic]

@Bjorn Gustavsson & @Image Analyst I am trying to divide the image (C2_down5um_246.jpg) into several parts like 1,2,3,4 and so on which indicated by test.jpg. Now, please help me to divide the image.

Thanks a lot

[Bjorn Gustavsson]

"indicated by test.jpg" is not specific enough for us to understand by what criteria you want your region boundaries to be determined.

Image Analyst below guesses that it is some gray-level intesity-boundary for the topmost region and then shifts it downward. I guessed that you could use some kind of functional form for the boundaries and get the intensities in the curved regions by interpolation.

You will have to understand that in order to get a solution that gives you precisely what you want you will have to precisely specify what it is you want. Your hand-scribbled boundaries can be used to generate boundaries but that is: a a hassle, and b seems rather imprecise and arbitrary.

Repeating your reference to the image with your scribbly lines will not give us any new and more precise information.

Answer 3

[Image Analyst]

Please attach your original image.

What I'd do is to try to get the upper surface/curve first since that is the brightest. You might be able to just threshold and then scan across finding the first pixel above the threshold. Then for the lower curves, just subtract some number from the top one. Something like

mask = grayImage > someValue;
mask = bwareafilt(mask, 1); % Take largest to try to get rid of noise.
[rows, columns] = size(mask);
% Find top curve first.
topRows = nan(1, columns);
for col = 1 : columns
    t = find(mask(:, col), 1, 'first')
    if ~isempty(t)
        topRows(col) = col;
    end
end
hold on;
% Make a bunch of them, each lower than the previous.
deltay = round(rows/8)
allCurves = zeros(7, columns);
for k = 1 : 7
    allCurves(k, :) = topRows + (k-1) * deltay;
    plot(allCurves(k, :), 'r-', 'LineWidth', 3)
end
% Clip to bottom of image
allCurves = min(allCurves, rows);

[Dynamic]

@Image Analyst & @Bjorn Gustavsson

My goal is split the image into small images which are in curved shape. So, my expected output for first shape indicated by 1 and second shaped represented by 2 and so on. Again I want to reconstruct my whole image. Please find the attached and expected output.

Thanks a lot for your help

[Dynamic]

@Image Analyst, Thanks for your support. Yes, I can add noise to whole image, however, I wanted to add different varience of noise for each band. Thanks for your code.

However, still, I have a question to sort out. If I want to 50% overlap band, then how to modify this code? Please, can you modified the code for that?

Thanks a lot