Researches On Image Compression And Denoising Based On Adaptive Bandelets Transform

Multiscale Geometric Analysis (MGA) is widely used in image processing such as image denoising, compression and enhancement. Utilizing the geometrical features of image can make the multiscale analysis method work more effectively. MGA is developed based on wavelets, which makes full use of geometrical features of an image , so it can represent the high-dimensional function effectively. The development of MGA tools provide us a new way for more efficient image processing. Bandelets transform is a MGA tool that represents images based on the edges, and Bandelets can follow the regularized geometric direction of the image adaptively, thus achiving a sparse image representation. This paper makes researches on image compression and denoising based on adaptive Bandelet transform. The primary work includs:(1) Combines the advantages of Multi-wavelet in the image compression such as orthogonality, approximately symmetry and compact support, at the same time a new geometric flow optimization strategy called collaborative decision geometric flow optimizatiuon is proposed based on multiscale sub_bands of Multi-wavelet, a Multi-Bandelets MGA tool is constructed based on Multi-wavelet and geometric flow optimization. It was used for SAR(Synthetic Aperture Radar) images compression, the results shows that Multi-Bandelets can obtain better performance both in visual fidelity and some objective evaluation criteria such as PSNR than its antitheses.(2) The geometric direction analysis (GDA) is introduced into Nonsubsampled wavelet transform, a Nonsubsampled Bandelet transform (NSBT) MGA tool is constructed. NSBT has advantages such as translation invariance and redundancy, it was used in image denoising and can obtained better performance than contrast method.(3) Utilize the dependence and relationship among the adjacent sub-bands of NSBT coefficients, the adjacent sub-bands are multiplied to construct NSBT multiscale product, it can enhance the geometrical feature of edge and restrain noise. It was used in image denoising, the threshold of NSBT multiscale product directly acts on multiscale product but not NSBT coefficients. This method is applied on the SAR image speckel reduction and good results are obtained than contrast methods. Researches on Image Compression and Denoising Based on Adaptive Bandelets Transform...