I have written a prototype for a project in C#, which in essence does the following:
Loads 2 images
For each pixel in image A find the corresponding pixel in image B (images are from stereo cameras)
Calculate the disparity between locations
Create a 3rd image from this.
It uses a fairly simple normalised correlation algorithm. However my problem is this. To perform this algorithm on every pixel of a 180x180 image takes around 5 minutes, whereas ideally it should take a few seconds (say 10 at most).
Im using standard C# (as in, just installed it without changing any options etc).
Are there any compiler options etc that are going to give me a major performance increase, or is a port to C++ / Java required? I have optimised the algorithm as much as possible (reducing search space etc).
Thanks for any help,