Parallel Research On Key Techniques Of Image Stitching Based On Hybrid Programming Model

Wide viewing angle and super-resolution image provide more accurate experimental data in many fields of scientific research, but the limits of image acquisition device and the immature of imaging technology cannot obtain satisfactory results, so the image mosaic become an effective way of solving the technology problem. However, the traditional image mosaic can only run on a single computer, and its performance is limited by the computer itself.The time of mosaic images spent will be increased rapidly because the images are too many.Although the image obtained could meet research requirements, we will lose the original image data timeliness. Based on this case, the thesis mainly studies research methods on SMP cluster MPI and OpenMP hybrid parallel computing parallelization of image mosaic, to accomplish the purpose of improved image mosaic speed. The main researches are as follows:Firstly, the thesis introduces the multi-core cluster parallel technology and MPI/OpenMP parallel programming model, the details of MPI and OpenMP parallel programming model and the implementation of parallel program. Then, it describes the steps of image mosaic, after that analyzes the process of extraction and matching algorithm based on the SIFT algorithm, and proposes a parallel algorithm for the image mosaic algorithm based on the SIFT algorithm.Secondly, according to the principle of the SIFT algorithm, the parallel algorithm is achieved in the data-parallel manner. In the light of SMP cluster hierarchical storage features, it is proposed that an image level partitioning strategy between nodes with different methods of image partition on different nodes to achieve corresponding particle size data distribution of MPI and OpenMP. It is proposed that the extraction and matching of SIFT feature points could be synchronous processed, and the feature points do not have to converge to the master node to be matched, but the process of matching will be completed in each sub-node, it will be sent to the master node to complete the image with the master node registration and fusion. In the data collection phase, the process of receiving feature points is simplified and parallel received.Finally, the image mosaic algorithm is based on the Hybrid Parallel Programming Model,and the DAWNING TC5000 is as the parallel environment. By comparing the parallel and serial experiments performed on the cluster, it is obtained the experimental data and the results of a detailed analysis. According to the experimental results of the analysis, the image mosaic algorithm which based on the Hybrid Parallel Programming Model greatly improved image mosaic speed. The results of image mosaic are consistent with that of the traditional one.