How to crop an image A which is a subset of an image B

I have two images where image A is a subset of image B.I could with feature matching technique - identify, match and show the images as a pair. Now that i have identified and matched image A with image B. i want to clip out(crop) that exact region from image B and create a new image. How can i do this, any help is appreciated ?

Answer 1

[Dima Lisin]

If you have matched the points, then you can simply find the bounding box of the matched points:

points = round(matchedTarget.Location);
left = min(points(:, 1));
right = max(points(:, 1));
top = min(points(:, 2));
bottom = max(points(:, 2));
croppedImage = target(top:bottom, left:right, :);

[dheeraj surabhi]

Thanks for your answer @ Dima Lisin

Can you please help me with this error :

>> croppedImage = Target(left:right, top:bottom, :);
Warning: Integer operands are required for colon operator when used as index 

Warning: Integer operands are required for colon operator when used as index 

Index exceeds matrix dimensions.

I tried using round() but still get the same error.

Thank You very much!

[Image Analyst]

Your left, right, top, and bottom are not row or column indexes. They are probably floating point numbers with fractional values. Obviously you can't have that.

[Dima Lisin]

Sorry, I've fixed the answer. Use round(), and switch top:bottom and left:right.

[dheeraj surabhi]

Thanks Dima. It works.

However i have a problem here. the fact that we are doing round() makes the cropped image loose a slight portion of edges as in the new image that i cropped out would have borders cut down slightly. This is happening since we are rounding the indices i guess, any idea on how can i overcome this and be able to crop the exact image?

Answer 2

[Thorsten]

If I understood correctly, you have found the part in image B that is exactly like image A, and now you want to crop that part of image B. Because both are identical, the cropped image would be exactly like A, so you can just write

C = A;

[dheeraj surabhi]

I guess what you meant is just to copy as a new image because they are identical anyway but assume that image B is true-color and image A is Gray. So i want to crop out the actual region from image B so that i can have a new color image that is identical to image A but is cropped out from image B.

[Thorsten]

 C = B(ind1,ind2,:);

where ind1, ind2 are the vectors ob indices that define the region.

Answer 3

[Image Analyst]

You have said you "have identified and matched image A with image B". If you did that, then you know the coordinates, for example like I did in my attached normxcorr2() demo. So simply use imcrop() after that.

[dheeraj surabhi]

I have identified using SURF feature matching technique but i don't know how to get the coordinates of the image A with respect to the image B. Here is my code where i could finally show the match as an overlay. If i could get that rectangle coordinates then imcrop would do the job.

% Detect features in both images.
ptsTarget  = detectSURFFeatures(Target);
ptsSample = detectSURFFeatures(Sample);

% Extract feature descriptors.
[featuresTarget,   validPtsTarget]  = extractFeatures(Target,  ptsTarget);
[featuresSample, validPtsSample]  = extractFeatures(Sample, ptsSample);

% Match features by using their descriptors.
indexPairs = matchFeatures(featuresTarget, featuresSample);

% Retrieve locations of corresponding points for each image.
matchedTarget  = validPtsTarget(indexPairs(:,1));
matchedSample = validPtsSample(indexPairs(:,2));

% Show point matches. Notice the presence of outliers.
figure;
showMatchedFeatures(Target,Sample,matchedTarget,matchedSample);
title('Matched points (including outliers)');

%Estimate Transformation
[tform, inlierSample, inlierTarget] = estimateGeometricTransform(...
    matchedSample, matchedTarget, 'similarity');

% Display matching point pairs used in the computation of the
% transformation matrix.
figure;
showMatchedFeatures(Target,Sample, inlierTarget, inlierSample);
title('Matching points (inliers only)');
legend('ptsTarget','ptsSample');

% Compute the inverse transformation matrix.
Tinv  = tform.invert.T;

% Compute the recovered scale and angle.
ss = Tinv(2,1);
sc = Tinv(1,1);
scale_recovered = sqrt(ss*ss + sc*sc) ;
theta_recovered = atan2(ss,sc)*180/pi ;

% Recover the indentified positions by transforming the sample image.
outputView = imref2d(size(Target));
recovered  = imwarp(Sample,tform,'OutputView',outputView);

% Show the sample image with recovered scale and angle as a pair blended
% into the Target image
figure, imshowpair(Target(:,:,1),recovered,'blend')

[Image Analyst]

Sorry - I don't have the Computer Vision System Toolbox. Perhaps Dima will answer you.