Super-resolution Multiple Frame Image Reconstruction Based On The Multi-core DSP

With the development of digital image processing, the actual demands for high resolution image are more and more urgent. The common imaging equipment can't meet the actual requirements. The difficulty and cost of equipment will be increased greatly,if the resolution is improved by hardware technology, such as reducing the pixel size or changing detection element arrangement. Image super-resolution reconstruction can solve the problem above-mentioned in a simple and effective way. However, its complexity so high that it can't meet the requirement of real-time. To overcome this problem, TMS320C6678 made by TI with 8 cores is used to complete image super-resolution reconstruction.The algorithms in spatial domain for the multi-frame image reconstruction are dissertated firstly in this thesis. Then the POCS(Projections Onto Convex Sets) is chosen for super-resolution by comparing different algorithms in spatial domain. Next the principle and process flow of POCS are researched and the complexity of POCS is analysised in depth. After analyzing the demand of super-resolution in this thesis, the programming architecture based on SYS/BIOS operation system is designed with parallel pipeline mode. According to the characteristics of parallel pipeline mode and POCS, load-balance is realized for each DSP core. Besides, the process flow of each DSP core is also designed. To maintain the Cache coherence, the dedicated memory is allocated in MSM(Multicore Shared Memory) for each DSP core. To achieve the communication between different cores, Heap Buf Mp is allocated for the MessageQ in each DSP core's memory of MSM, and then the data can be transmitted successfully with Message Q_put() and Message Q_get(). At last, the POCS for multicore mapping is achieved with the CCS(Code Composer Studio) and corresponding SDK(Software Development Kit).The verification results of demonstration platform showed that all test results can well meet the requirement of super-resolution in this thesis. On the other hand, the speedup ratio of parallelization computing can reach up to 6.254 in the case of eight processor cores' working. The results showed that multi-frame image superresolution reconstruction based on multi-core DSP can improve the real-time effectively. Besides, it can also reduce the cost of imaging equipment greatly.