I am trying to create a image text detection program. The program is up and running, and I can distinguish between text and non-text using laplacian based method with an acceptable accuracy.
However, the problem is that I need the program run with the speed > 100 (images/s), and the best record so far is 10(images/s) with the resolution of 1280x768. My program spend most of the time on Laplacian convolution calculation and maximum laplacian difference.
I tried most of the optimization methods that I have heard of. The only thing that I haven't tried is programming in assembly code due to the complexity of my program.
Any suggestion will be appericiate.