FPGA structures with concentrated vs distributed memory for images comparison

This paper proposes a strategy for calculating the windowed correlation between histograms using FPGA. The processing is done extracting image features directly from the compressed video stream. Thus, maximum speed is obtained; the circuit can manage up to 200.000 photograms per second. Two options for calculating the histogr am are proposed: one based on distributed registers, and other based on blocks of RAM. The performance of the two a Iternatives is compared. Exhaustive registers-level segmentation for the algorithms and block memory-level segmentation for histograms is applied. The final circuit is evaluated taking into account its adaptation to different video resolutions, number of histogram bins, parallelism in the concurrent computation, and progress of FPGA technology.