Fast Image Cartoon And Texture Decompositon Based On Sparse Represention

With the rapid development of multimedia technology, digital image processing technology have had a great development in people’s life, image denoising, image edge detection, as well as involved in the computer vision many new algorithms are emerging. As a new attempt, image cartoon and texture decomposition can overcome the shortcoming of the traditional image edge detection which can only be used to detect the global oscillation in image but can not portray overall outline of the image. In the high level processing, such as the computer vision the decomposition parts can be used in target identification, so the image decomposition in image processing area will get more applications.This article introduced the widely used sparse representation of image into image cartoon and texture decomposition, classify the natural images which contain texture features and cartoon features, for the shortcoming of introduce interference patch in the image random sampling process, in this paper we use the K-means cluster strategy to classify the sampled image patchs, then we trained the image patches to get the cartoon dictionary and the texture dictionary. By seeking the sparse representation coefficient of natural images which is in the context of cartoon and texture dictionary, and find the sparse representation coefficient which corresponding to the texture and cartoon dictionary then will be able to achieve the image cartoon and texture decompositon. However in the decomposition process have to compute the sparse representation coefficient of every patch, the time it takes in the optimization process is greatly increased, so the time of the decomposition of the whole image is very long. So in order to overcome the shortcomings of having to compute the sparse representation coefficient of every patch. In this paper, a new training dictionary stategy for image super-resolution algorithm is introduced to the work of image decomposition,the process of computing sparse repersentation coefficient of every patch can be transformed into the simple process which we have only to use the patch multiply the corresponding dual dictionary. Thus greatly reducing the computational time. In the analysis of the experimental results we can see that the method based on sparse repersentation and the PADDLE fast algrithm mentioned in this paper compare with the nonlinear filter and the MOM which based on MCA,our method can keep the overall edage better in the cartoon component after image decomposition,and can overcome the shortcoming of texture information remove incomplete which happened in the method of MOM and nonlinear filter.During the compare of compute speed our method is less rapaid than nonliner filer, enhance about20times compare with MOM and the sparse repersentation method.Our decomposition method can achieve good results both in the decomposition effect and the decomposition speed.