Sparsity Adaptive Matching Algorithm And Its Application In Face Recognition Tracking

Traditional Nyquist sampling theorem must be satisfied that sampling frequency is more than twice the maximum frequency in order to remain the important information. However, with the increasing of bandwith sampling data grows in quantity, the sampling theorem bring more difficulty to sample, process, merge and transfer the information. Compressive Sensing (abbrevieated as CS) theory has broken the limit of the sampling theorem, it combines sampling with compressing if the signal is compressive or sparse. Therefore, more and more researchers pay attention on the CS in the signal processing. And signal reconstruction is significant to verify the accuracy of the sparse representation and measurement in the CS.This dissertation firstly discusses some kind of signal reconstruction algorithm, and focus on matching pursuit algorithm. The main contributions and research of the thesis are as follows:Firstly it introduces a few common matching pursuit algorithm, for example Orthogonal Matching Pursuit (OMP), Regularized Orthogonal Matching Pursuit (ROMP) and Sparsity Adaptive Matching Pursuit (SAMP).In addition, they reconstruct the gray image and the experiment result indicates that the performance of SAMP is better. But, the fixed-step approach to the sparse K in the SAMP, which may cause under-estimated or over-estimated. Therefore, the dissertation improves the SAMP algorithm in order to solve this problem. According the variation of the energy difference between adjacent signal, it uses logarithmic "variable step" in the iteration, that is it uses "big step" which means step size increase quickly in the initial stage, and then it uses "small step" which means step size increase slowly, as well as dual-threshold for accuary. Experiments show that improved SAMP not only ensure the quality of reconstruction, but also reduce the iterations time.Moreover, Sparse Repressentation-based Classificaiton is an effective solution to the curse of dimensionality of face recognition, at the same time it can get higher recognition rate even though in low-dimensional feature space. Therefore the dissertation combines the improved SAMP with Sparse Repressentation-based Classificaiton, the experiments based on different face databases show that it has better robustness and higher recognition rate.