Natural image noise removal method based on dual redundant dictionary learning

Abstract

The invention discloses a natural image noise removal method based on dual redundant dictionary learning, aiming at mainly solving the problem that texture details are easy to lose in the existing natural image noise removal method. The natural image noise removal method based on dual redundant dictionary learning comprises the following steps: (1) inputting an image Y to be subjected to noise removal; (2) unfolding the image Y to be subjected to noise removal under a multi-scale redundant stable wavelet dictionary R to obtain coefficient components betaj of the image Y on different scales, wherein j is 1, 2,... and N; (3) initializing each variable to make a dictionary Dj be a redundant dictionary DCT, and updating atoms and corresponding coefficient matrixes of the dictionary Dj by adopting the KSVD algorithm; (4) calculating the estimation value of the noise removal results of all the coefficient components betaj; and (5) carrying out multi-scale redundant reverse transformation on all the coefficient components betaj subjected to noise removal to obtain the noise removal result of the image Y to be subjected to noise removal. Compared with the existing classical noise removal method, the natural image noise removal method based on dual redundant dictionary learning can better keep the texture detail information in the image to be subjected to noise removal and can be used for natural image noise removal.

Description

technical field [0001] The invention belongs to the technical field of image processing, in particular to a denoising processing of natural images, which can be used in the fields of image processing, pattern recognition and biomedicine. Background technique [0002] The goal of denoising is to preserve the feature information of the image, such as texture, edges and point objects, while removing noise. The noise of natural images has an additive background, and denoising can be carried out from both the spatial domain and the transform domain. Typical spatial domain filtering methods include Lee filtering, non-local mean filtering, and redundant dictionary image denoising. in: [0003] Lee filtering adopts the mean value in the homogeneous area, and adopts the local filtering strategy for the fast-changing points. One of its disadvantages is that it cannot effectively remove the noise around the edge or the texture of the over-smoothed image; the second disadvantage is: O...

AI Technical Summary

Problems solved by technology

Its shortcoming is that the atoms of the DCT dictionary are fixed, which cannot effectively approach the edge and detail information of the original image, and the error control method of KSVD dictionary learning is rough, which may easily cause the loss of some texture details in the original image.

Images