Regularized parameter adaptive sparse representation image reconstruction method

Abstract

The invention provides a regularized parameter adaptive sparse representation image reconstruction method, comprising the following steps: adopting a sparse dictionary learning algorithm, extracting image edge features to train a compact sparse dictionary, and adaptively allocating sub-dictionaries to each image block for sparse coding; Learning the Sparse Estimation of Sparse Coding from an Overcomplete Dictionary; Making full use of the local structure similarity of images, the regularization parameters are adaptively solved by using the method based on the maximum a posteriori probability.Adaptive sparse representation model of regularized parameters is established. The invention improves the effectiveness of sparse representation by utilizing the local structure similarity, and the edge and the structure of the image are well maintained. Adaptively adjusting the regularization parameters based on the maximum a posteriori probability, updating the regularization parameters in eachiteration process, better adapting to the current situation, greatly reducing the workload of manual selection of the regularization parameters; It has good image reconstruction effect and strong robustness to noise and motion blur.

Description

technical field [0001] The invention belongs to the technical field of image processing, and relates to a super-resolution image reconstruction method, in particular to a regularization parameter self-adaptive sparse representation image reconstruction method. Background technique [0002] Image super-resolution (referred to as SR) reconstruction is to reconstruct a high-resolution image by processing one or more low-resolution images with complementary information. Due to the ill-posed nature of the super-resolution image reconstruction problem, regularization-based methods are widely used to solve this ill-posed problem by regularizing the solution space. In order to obtain effective regularization terms, prior knowledge suitable for natural images should be found and established, and various image prior models have been researched and developed. Classical regularization models, such as reconstruction algorithms based on Tikhonov regularization and TV regularization, can ...

AI Technical Summary

Problems solved by technology

[0003] However, the super-resolution reconstruction algorithm based on sparse representation is prone to excessive smoothing when image reconstruction is performed, loss of detail information, and the edge and structure of the image cannot be completely maintained.

However, traditional super-resolution image reconstruction methods, such as bilinear interpolation, TV regularized image reconstruction, and image reconstruction based on sparse representation, are less robust to noise and motion blur.

Images