The Parallel Image Matching Algorithms Research In Chip Multi-Processors

Currently, Chip Multi-Processors(CMP) is one of the processor's development trends. In this architecture, we can easily impolder the thread-Level parallelism. The image matching technology are widely used in both military and civilian domains, it relates to many image processing algorithms. Because of the digital image has geometrical parallelism naturally, it can be processed in parallel easily. The CMP architecture have two bottlenecks, viz. the I/O channels bandwidth and the local memory capability are limited. In addition, the edges of image can not be processed when it was processed in parallel. So, designing effective parallel image processing algorithms is very important.In this thesis, we showed the processor's development trends and introduced the characteristics of CMP architecture. We described the principle and method of image matching. Then, we analysed the filtering, edge detecting, distance transforming and sorting algorithms, according to the characteristics such as data dependency, we presented three parallel algorithms. These algorithms can break through the two bottlenecks of CMP architecture effectively, and they can achieved to high speedup values. We implemented the three parallel algorithms with MPI(message passing interface) to verify their correctness. The presented parallel algorithms can be refered in designing CMP system and optimizing other applications.At last, we designed a image matching system, it integrated many image processing functions. The result of matching is very accurate when using this system.