compv
compv copied to clipboard
Sobel could be done multiplication-free
Sobel kernels are pre-defined (0, 1 and 2 values) which means we can compute the convolution without using multiplications (add and sub only). For now sobel convolution not used in commercial products which means not urgent to change the implementaion.
And use abs_avg ((abs(gx) + abs(gy) + 1) >> 1) instead of gradient L1 plus scaleAndClip. abs_avg is single instruction in SSE and NEON