Efficient Algorithms For Image Restoration And Parallel Processing

In the practical transmission of block-based coding image, bit errors may occur, which often results in lost block. Error concealment is a widely used technique to conquer the transmission error. It has been proved that Best Neighborhood Matching (BNM) algorithm is a very good error concealment approach in terms of restored image quality [1]. However, lots of error concealment algorithms (including BNM) are computation intensive, which restricts their real applications on large-scale image restoration or image sequence restoration. In this thesis, we propose a fast image restoration algorithm to speedup BNM.Parallel processing is an importance way to deal with problems with large computation. However, in parallel system, the development of computing ability, communication speed and I/O processing ablity is not proportion, which makes it is not an easy thing to design efficient parallel algorithms. In this thesis, we will study the important issues and the optimal techniques, which influence the efficiency of parallel algorithms. In particular, we will study the isssues, which infuence the I/O performance, and develop high performace I/O strategy. Meanwhile, a high efficient parallel algorithm for image restoration is developed.The achievements of this thesis can be summaried as follow:(1). A fast image restoration approach, named JLBNM (Jump and Look all round BNM), is proposed to conquer the limitation of BNM. The main idea of JLBNM is based on the fast searching method of eyes' intelligence of human being. Meanwhile, some important optimal techniques are developed, including the adaptive threshold in the matching stage, the terminal threshold in the searching stage, and the right matching function in both the matching and recovering stages. From our theoretical analysis and simulation results, it shows that JLBNM not only provides a good block-based image restoration approach with the high quality, but also has the low computation complexity.