I'm in the planning phase of a task where large images are to be edge enhanced. The filter in use would be reasonably simple, such as Laplacian of Gaussian (LoG) or similar. The images would be say 2500x2500 pixels, 8-bit grayscale.
Before spending a few days writing test code I'd like to get a rough idea of what kind of performance one would expect for this operation from IPP? Ignoring the time required to allocate source and destination images/areas, what order of magnitude (ms) would the actual processing take?
Any guesstimates would be greatly appreciated!
Also, would it be possible to "link" filters, or would I need intermediate images/filters if gauss and laplace filters are to be executed in sequence? I guess the alternative would be to use convolution with the corresponding composite filter kernel.