Reduced processing element architecture for parallel local image processing

The object of this paper is to propose new architecture which can reduce the number of processing elements for parallel local image processing under the premise of real-time performance. For large-sized local image processing, this architecture will save much space as it is suitable for being designed into VLSI chip. For example, the traditional parallel architecture will use 9 PEs for a 3X3 convolution, while the Reduced Processing Element Architecture (RPEA) only requires 2 PEs to achieve the real-time performance.