Periodic Noise Removal Based On Bandelets

Image tends to be affected by noise in the process of acquiring, transmitting and processing. On the one hand, image quality is reduced by the noise. On the other hand, a lot of information carried by image is destroyed. As a result, the image denoising is particularly important.In the digital images, the noise in image is produced in the process of acquisition and transmission. As a special kind of noise, the periodic noise is mainly produced by electrical or electromechanical disturbance during the acquisition of the image, which is a kind of spatial correlated noise. The periodic noise can be regarded as a texture attached to the image. The morphological component analysis is a kind of image separation technology. Cartoon and texture separation is a special form of the morphological component analysis, which can make the cartoon separated from the texture. This paper firstly studies some of the tools used in image analysis and presentation and then studies the morphological component analysis and the algorithm of MCA. The algorithm of cartoon-texture separation is improved. In the selection of the dictionary, the second generation Bandelet transform is improved, and the improved Bandelet transform is selected as the cartoon dictionary. The periodic noise is removed through the method of cartoon-texture separation.The main work of this paper is as follows:Firstly, the morphological components analysis based on sparse decomposition is applied to the removal of periodic noise. The cartoon-texture separation algorithm selected Curvelet as its cartoon dictionary is improved. Compared with the original algorithm, the algorithm improved improves the effectiveness of the cartoon-texture separation.Secondly, Bandelet transform is studied in depth. The character that Bandelet transform can represents geometrically regular image adaptively is used fully. The objective function used to selecting the bases of Bandelets is defined. At the same time, the cartoon-texture separation algorithm selected Bandelet transform as cartoon dictionary is put forward and applied to remove the periodic and quasi-periodic noise. Experimental investigation shows that the method based on Bandelets can remove the periodic and quasi-periodic noiseadaptively. Especially, the effectiveness of removing quasi-periodic noise is better than the method of frequency domain.Thirdly, the periodic noise denoising method based on Bandelets presented in this paper is applied in remote sensing image. The experimental results show that the method eliminates the periodic strip noise in the remote sensing image adaptively while preserving original information of the image.