How to choose the fixed points in fitgeotrans method for image rectification

Hello, help me please understand the fitgeotrans funcion in Matlab.

I have different images of various sizes of a square object, and in different perspective views; in all of this images I have detected the object and found its 4 corners. Now what I want to do is to make a perspective rectification of these images so that after I can get a "plan view" of this object and then crop the image around this object.

I have implemented this code in Matlab based on some search in the Web:

movingPoints = corners; %contains the 4 corners of the object in clockwise order
fixedPoints = [0 0; 500 0; 500 500; 0 500];
tform = fitgeotrans(movingPoints,fixedPoints,'projective');
warp = imwarp(image,tform);

And it's working, I get my object in "plan view". However the object rectified sometimes is very big(or near), and sometimes is small(or distant can we say); I suppose it's because of how much "work" has done the transformation.

How do I choose the fixedPoints so that after I can crop my object knowing the new corners position in the image rectified ?

Answer 1

[Diwakar Ravichandran]

Hi,

As I try to understand your problem, there remains a few things I am not yet clear on. So, fitgeotrans is a function that can be used to fix distortions in your images based on the fixed points and the moving points. As shown in the documentation, https://www.mathworks.com/help/images/ref/fitgeotrans.html

The method of choosing your fixedPoints is based on the non distorted image. So when you have a non distorted image, and you identify specific points on the image, and these become your fixed points. And the rest of your distorted images uses these fixed points to generate a transform to help your moving points to reach their fixed points.

Hope this helps,

Cheers!

[Diwakar Ravichandran]

One other thing to also have a look at

https://www.mathworks.com/matlabcentral/answers/290703-how-do-i-apply-a-fitgeotrans-output-or-similar-to-points-not-an-image?s_tid=answers_rc1-1_p1_MLT