Study On Algorithms For Image Processing Based On Directional Mutiresolution Analysis

Contourlet transform employs an efficient tree-structured implementation, which combines Laplacian Pyramid transform and Directional Filter Banks. The redundancy rate of Contourlets is up to 4/3, and it is directional and multiresolutional but is not shift-invariant transform. The Non-subsampled Contourlet transform is combined with non-subsampled Laplacian Pyramid transform and non-subsampled Directional Filter Banks. It is multidirectional and multiresolutional, and it is shift-invariant. But it is highly redundant and computationally expensive.Surfacelet transform is a true 3D wavelet transform. It offers a directional multiresolution decomposition of 3D signals with a redundancy ratio of up to 24/7. It provides more sparsity of subband than the wavelet.This paper has two main contributions. Firstly we propose Analytic Signal Contourlet transform and apply it to Context Based Image Retrieval. The Analytic Signal Contourlets can suppress the aliasing effect, and the experiments show that the new method significantly improves retrieval rates, e.g. from 76.39 to 80.45, compared with the method from [32] while it is effective implementation.Secondly we introduce the theory of Surfacelet transform and apply it to video watermark exploring its good directional decomposition. The experimental results show the proposed watermarking algorithm is more robust to video degradation and distortions, e.g., additive Gaussian noise, MPEG coding at high compression-ration compared with the method from[72].