Could anyone help me with extracting similar features( to be used as corresponding points)in two images which need to be registered together?

Hello all, I am trying to register the two attached images together, for this purpose I need corresponding similar points in the two images. For extracting such points, I tried to use connected components, regionprops, imfill, imclose functions but I couldn't get similar features. Could you please guide me if you have ideas about this problem. Thanks so much.

[Anton Semechko]

Is there a specific reason you want to use a point-based vs. intensity-based registration? The latter can be performed without extraction of corresponding point sets from the images, which is a non-trivial problem in and of itself.

[Penny13]

Actually,This is my first time doing a registration and I am not very familiar with all methods deeply but the thing is being able to automatically extract corresponding points and use them in registration is the important step in this task that I have been asked to perform.

[Anton Semechko]

OK, so the reason you want to do point-based registration is because you have been asked to do so. In my opinion, it will be much easier to align these images using an intensity-based registration approach. It will also be much easier to identify corresponding point sets in the images after they have been aligned.

[Penny13]

Thank you so much for your answer. I appreciate that. I will try the intensity based registration but also I need to extract similar features too.

Answer 1

[Anton Semechko]

Hey, Poupack,

here is link to a function ('pairwise_histology_registration.m') that performs point-based registration of two histology images. This function detects corresponding point sets in the images using SURFs ( Speed Up Robust Features ). You can find additional info on these and other features implemented in the Computer Vision System Toolbox here.

A quick demo using your sample images:

% Sample histology images

im1=imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/124273/1.jpg');
im2=imread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/124274/2.jpg');

% Register im2 to im1 using 'pairwise_histology_registration' function. 
% In medical image processing jargon, im1 is called a reference image, and 
% im2 a source image. Whether im1 or im2 is designated as a reference is 
% not important. See function documentation for additional info.

[im_reg,T,RMSE,SD]=pairwise_histology_registration(im1,im2);

In this example, a total of 1351 corresponding point pairs were detected and aligned with accuracy of 0.5 pixels using a rigid transformation model; top 200 strongest matches shown in green. The function also supports similarity and affine transformation models. All relevant info about corresponding point sets detected by the function is contained in the structure represented by the 'SD' variable.

---

reference image

---

source image

---

source registered to reference

[Penny13]

Thanks a lot Anton for your nice job and guiding me toward this problem.

[Anton Semechko]

Glad to help. Let me know if you run into any problems using this function.

The two sample images you provided were very similar and mostly differed in terms of translation; enabling registration to be performed with sub-pixel accuracy. I am curious to know how well the function performs on more challenging cases. For example, when histology slices only have partially overlapping structural content or nonlinear deformations.

[Penny13]

Sure. Thank you so much. Likewise, If I face such images I will send you the images in case you like to see the results.

[Anton Semechko]

That would be great, Poupack. Thanks!

Answer 2

[Penny13]

Sure Anton!