Research Of Visual Information Parallel Processing Algorithm Based On Reconfigurable Processor

Visual information supplied more and more, facing real-time requirements. Itrequires a high speed computing architecture to solve the deficiencies of the existinggeneral-purpose processor and a special processor. A reconfigurable processorcontains a general-purpose processor and a reconfigurable cell array, havingflexibility, efficient, and data parallelism. Its reconfigurable characteristics canoptimize the efficiency of visual information processing algorithms. This dissertationstudies parallel visual information processing algorithms based on reconfigurableprocessor, the main research:(1) Analysis of visual information processing in the image mosaic algorithm basedon feature, presents a normalized cross-correlation algorithm based on gradient. Therelative gradient is utilized instead of the intensity information in Normalized CrossCorrelation (NCC) measure.(2) Describes reconfigurable processor hardware architecture, and according to thedistribution of running time of the algorithm, fully exploit the parallelism of thealgorithm the most concentration computation. Break the algorithm into operators,form an operator library based on reconfigurable processor.(3) Use threads simulate reconfigurable array element (PE), set up multiplethreads parallel execution on multi-core computer by calling OpenMP, which cansimulate the calculation process on reconfigurable processor platform. With thisapproach, the algorithm is mapped to a multi-core computer through the use ofnuclear simulate PE unit, in the absence of a specific program of PE connection. Thus,execution efficiency of visual information algorithms can be analysis on multicorecomputers to find the optimal mapping programs on reconfigurable processing.The improved algorithm is effective to reduce the feature point matching rate bymistake. The simulation results show the validity and reliability of parallel operator.