Image Denoising Algorithm Based On New Multiscale Transform

With the increasing popularity of computer and the rapid development of Internet,we are free from the way which we pass on a message with letters or telegram. Through the image to get the information in the real world is becoming more and more popular. Noise can be understood as to interfere with human visual sensor or system accepted source of information for understanding the various factors, the image in the process of acquisition, acquisition and transmission, tend to be polluted by noise.Noise not only affects the resolution of the image, more important it will affect us for accuracy and completeness of information. It is for this reason, The image denoising become an integral part of the scientific research field. The success of image denoising will seriously affect the subsequent image processing operations.In this paper, new multiscale transform is the main line, the two image denoising scheme is put forward:1.Using support vector machine(SVM), coefficient can be classified, this paper proposes a image denoising algorithm based on a proximal classifier with consistency(PCC).The PCC is improved SVM algorithm, improves the classification accuracy and speed. First of all, nonsubsampled Shearlet transform resolve the image into different scales of subband.Secondly, using the PCC classify the nonsubsampled Shearlet transform coefficients into noise coefficients and the non-noise coefficients; Then, noise coefficients are filterd directly, the non-noise coefficient of adaptive threshold denoising processing; Finally, throuth the nonsubsampled Shearlet inverse transformation after denoising image. This scheme can output higher PSNR under the condition of the smaller relative error.2. In recent years, the HMT model is widely used in the field of image processing, there are also many models were constructed for image denoising processing. In this paper, based on the new type of HMT, on the basis of HMT model before increase the relationship between the coefficient of scale, can better describe the factor structure, and then to deal with the noise. First of all, the Bidimensional empirical mode decomposition(BEMD) resolve the image into different scales of subband and using distribution index count conditional probability density;Second, the calculation of the multi-dimension average taper ratio, and the joint probability density of new type of HMT training; Finally, the application of bayesian shrinkage function to deal with the noise. By reconstructing image can be seen that the solution can keep image information.