I implemented a function that makes use of OpenCV's estimateRigitTransform(). I store the result in a OpenCV Mat called T:
Mat T = cv::estimateRigitTransform(image, reference, false);
I know that by setting the last argument to false I get a partial Affine transformation that combines rotation, scaling, translation and has 4 degrees of freedom. ('true' would output a 6 degrees of freedom fullAffine transformation that includes shearing.)
When I print the Mat T, I get the following result:
[1.050438, 0.044571, -37.17997;
0.059126, 0.952621, -3.825856]
I have problems interpreting this transformation matrix. From what I understand the matrix represents the following:
[cos(theta)s, -sin(theta)s, tx;
sin(theta)s, cos(theta)s, ty]
which is a combination of a rotation matrix cos() and sin(), a scaling matrix s and a translation vector tx and ty.
So as the scaling factor s and the angle theta are constant, shouldn't the two cos(theta)s values be the same? In my out output one is 1.050438 and the other 0.952621?
When I display the image transformation using OpenCV's wrapAffine() and passing that matrix T, the output seems quite reasonable but beside of that I have no possibility the check T for correctness.