I'm trying to use IppiIntegral_8u_32f(...) andIppiIntegral_8u_32s(...) functions to create an integral image from my original image (one frame from video) in grayscale. It works fine, but I have two questions:
1.) My image height is 288 px and width is 384 px. When I trying to use ROI larger then h: 288 and w: 353 [px] the program crashes (Windows error: "Program can't continue working. Please contact with your reseller..."). I suspect that it's connected with memory exceed, but I don't understand why two types of IntegralImage function (based on float and signed int) have the same error boundaries (353 px of width and 288 px of height)? I know that float type can contains number as big as 3.402E+38. I found a table in Intel documentation, where are presented boundaries for these functions, but e.g for IppiIntegral_8u_32s it's (2^31-1)/255=8421504,498, BUT my images has 384*288=110592 px size, so why I can't create integral image from whole data?
2.) pSrc parameter is described in documentation as "Distance in bytes between starts of consecutive lines in the source image.". If ROI is less than image, if I have to put here image width*number_of_bytes or maybe ROI width*number_of_bytes?
Thanks for help.
it is important to notice that destination image for ippiIntegral function requires more memory (basically itshould bewidth+1 and height+1 in size)
Regarding ROI please refer to IPP reference manual, Image and Video Processing, IPP Concepts, Major Operation Models, Region of Interest in Intel IPP.