Re: Fixing bug in Harris/Hessian examples

From: Olzhas Zhumabek (anonymous.from.applecity_at_[hidden])
Date: 2019-09-16 17:23:23

After thinking about it a bit, it seems like only first and second order
derivatives are of interest, and higher order ones are very rare. My plan
at the moment is to add generate_dx_sobel/scharr(unsigned int power) and
generate_dy_sobel/scharr(unsigned int power) that will generate Sobel and
Scharr kernels for horizontal and vertical directions respectively. There
is algorithm to compute ith power, but it requires 2d convolution with
mirroring edge values. I guess I could just write a switch statement,
returning predefined values for 0th ([1]), 1st (from reliable source), 2nd
(from reliable source), and then leaving out the other powers for now.
There is a function that Pranam implemented, but it is incompatible because
there is extension only by two rows and columns, not sure if such mirroring
is defined for larger values.

What do you think?