I am trying to evaluate the correct implementation of a SURF descriptor with a basic structure from Mikolajczyk et. al . I use OpenCV to discover and describe SURF functions and use the same function positions as the input for my handle implementation.
To evaluate the performance of the descriptor, the structure must first evaluate the repeatability of the detector. Unfortunately, the repeatability test involves a list of function positions along with ellipse parameters that determine the size and orientation of the image area around each function. However, the OpenCV SURF detector only provides function position, scale, and orientation.
A related article proposes to calculate these ellipse parameters iteratively from the eigenvalues of the second moment matrix. Is this the only way? As far as I can tell, this will require some mess with OpenCV. Is it impossible to subsequently calculate these parameters of the ellipse (for example, in Matlab) from the list of functions and the input image?
Has anyone ever worked with this structure and could help me with some insights or pointers?
source
share