Research On Image De-noising Algorithm Based On Low Coherence Dictionary Learning

In recent years,the application of dictionary learning for image de-noising has attracted the attention of many scholars.The basic process of dictionary learning is mainly to learn the given sample set by adaptive learning method to get the optimal atomic set,and then use less atomic The linear combination is used to sparsely represent training samples.Commonly used dictionary learning methods include DCT(Discrete Cosine Transform)algorithm,MOD(Method of Optimal Directions)dictionary learning algorithm,and K-SVD algorithm.However,they do not consider the dictionary.The important problem of coherence between atoms leads to the high similarity between the noise atoms in the dictionary and the noise-free atoms,which makes the image de-noising effect worse.To solve this problem,this paper constructs a kind of reduction.Dictionary sacristy can effectively reduce the model of dictionary coherence,and two different algorithms are proposed for the model.The specific algorithm is as follows:Method 1.In order to reduce dictionary coherence and speed up the sparse coding,in the dictionary update phase,the singular value decomposition method is adopted,but the eigenvector corresponding to the largest singular value in the K-SVD algorithm is no longer used to update the dictionary,but instead The eigenvectors that can minimize the model are used to update the dictionary.The learned dictionary is used for image de-noising.Numerical experiments show that the improved algorithm has less coherence than K-SVD and obtains better de-noising effect.Method 2.Based on the objective function established by the above model,the quasi-Newton gradient tracking algorithm is used in the sparse coding phase to replace the OMP algorithm to solve the dictionary sparse coefficient.In the dictionary update phase,the dictionary in the objective function is graded to make the gradient equal to zero.The iterative formula of the dictionary is obtained,and then the dictionary is solved.The experimental results show that the model established in this paper has higher robustness.