| The further research of the Compressed Sensing(CS) theory draws many researchers attention in signal and image processing area. Natural images are proven has the feature of Sparse Representation, which accord with much signal and image processing work people usually contact. As the improving of Sparse Representation algorithms, especially after put forward of learning dictionary theory, the CS theory has been applied to signal and image processing fields.Based on the CS theory, and centered on image recognition and super resolution reconstruction of images, this thesis focuses on how to construct a classification recognition algorithm with high robust and low computational complexity, and how to construct overcomplete dictionary to build super-resolution image more effectively.The main works are described as follows:1) Based on Sparse Representation, SRC (Sparse Representation-based Classification) theory framework is stated in details. Due to the high computational complexity in using the L1 norm minimize algorithm to get the Sparse Representation coefficient of SRC methods. We propose an OMP (Orthogonal Matching Pursuit) algorithm with lower computational complexity to improve SRC method.2) For the problem of the low recognition rate of SRC method in under low dimensional case, a new SRC recognition method based on the categories information is presented. Face recognition and space target recognition experiment shows that this method improves recognition rate under low dimension case and has strong anti-noise ability.3) Based on Sparse Representation, image reconstruction algorithm is stated to reconstruct super-resolution by using YCbCr or RGB color image model, and also contrasted to traditional methods such as nearest neighbor interpolation, wavelet interpolation and etc, and the experience result shows that image reconstruction algorithm based on Sparse Representation has obvious advantages compared with traditional method on reconstruction of grey level and color images. |
PDF Full Text Request