Research On Image Denoising Based On Dictionary Learning Algorithm

Image plays an important role in human life, but due to various factors, the image quality is not good, the acquared image is usually polluted by some noise, which influences the visual effect. Image denoising is developed based on this problem, the purpose is to preserve the information of important image features, such as edge and texture, while reducing noise.Recently, redundant sparse representation theory based image denoising algorithm has gained wide attention and study. The algorithm uses the sparsity of image signal to distinguish signal and noise, and then realizes image denoising. Based on the research of sparse representation and redundant dictionary learning based image denoising algorithms, we discuss existing redundant dictionary learning methods, and then propose two new image denoising algorithms. The main contents of this paper are summarized as follows:1. An image denoising algorithm based on cluster ensemble and sparse dictionary learning is proposed. Regions with similar structure are obtained by steering kernel regression weight with cluster ensemble, by the idea of nonlocal. Then the sparse dictionary learning algorithm is applied to these regions of same class. The algorithm can overcome several shortcomings of traditional K-SVD algorithm, such as the structure of the dictionary is not strong enough, and the self-similarity of image is ignored. Experiments on natural and medical images validate the effectiveness of the proposed algorithm. The proposed algorithm can not only reduce the noise, but also improve the smoothness of homogeneous regions and preserve detailed features of image.2. An image denoising algorithm based on improved K-SVD dictionary learning and similarity constraint is proposed. Dictionary update and sparse coding in the K-SVD algorithm are improved in the improved K-SVD algorithm. Then a similarity constraint term is introduced into the objective function of improved K-SVD algorithm and get a new objective function. Experiments on natural and medical images demonstrate the effectiveness of the proposed algorithm. Compared to the algorithm in section 3, theproposed algorithm in this section exceeds in both visual effect and numerical index.