General 2D hough transform

Memory intensive but fast - contains no for loops. Best suited for non-binary inputs resulting in weighted results, but works just fine for binary images as well - though perhaps slower than needed since a multiplication is probably slower than an 'and' operation.
Two inputs required:
1) The input image to have the hough performed upon (IJ)
2) The kernel image (k). This can be any image - a line, circle, wrench... It doesn't matter as long as it is greyscale and positive. The output marks are with respect to the center of the kernel.
Probably works best for pixel values ranging between 0 and 1, but this isn't likely to be a big deal.

Assumes background is zero.

Best suited for sparse images - the fewer non-zero pixels the more this outperforms other Hough transform functions. So, anything you can set to zero is a good idea, but still works fine if you can't afford to (assuming you have the memory).