An image compressed sensing reconstruction method based on a non-local self-similarity model

Abstract

The invention requests to protect an image compressed sensing reconstruction method based on a non-local self-similarity model, and belongs to the field of signal and image processing. Specifically, in order to improve the quality of traditional image compressed sensing reconstruction, prior information of the image is utilized; a non-local self-similarity model of the image is constructed; a weight matrix of an image block is calculated; constructing an adaptive non-local regularization item of the image by utilizing the non-local self-similarity priori information of the image block; a mathematical model of image compressed sensing reconstruction is provided, and an efficient Spant Bregman Iteration (SBI) algorithm is used for alternate iteration updating, so that the reconstruction performance of image compressed sensing is improved. Meanwhile, in the learning process of the dictionary, a training sample is extracted through a current approximate estimation image d, and K-is utilized; and the SVD algorithm is alternately updated to obtain a self-adaptive learning dictionary. The adaptive image compressed sensing reconstruction method of the non-local self-similarity model provided by the invention has universal significance in practice. The image compressed sensing reconstruction quality is effectively improved, the block effect of the image is reduced, the texture and details of the image are kept not lost, and the texture and details of the image are better depicted.

Description

technical field [0001] The invention belongs to the field of signal and image processing, in particular to an image compression sensing reconstruction method based on a non-local self-similar model. Background technique [0002] With the continuous progress of society, information and its acquisition methods have become an indispensable part of people's daily life. The rich information content of images makes them the main information source for human beings to obtain information. However, the amount of digitally processed image data is very large, which brings considerable difficulties to the actual storage, transmission and understanding. At the same time, due to the limitation of hardware and the cost consideration of improving transmission efficiency, it is very difficult to achieve the goal with very little data. Representing, retransmitting and storing important information becomes the key to solving these problems. The establishment of image sparse representation mod...

AI Technical Summary

Problems solved by technology

However, there are two defects in this traditional compression method: 1) due to the limitation of the Nyquist sampling theorem, the sampling rate of the signal is higher than twice the signal bandwidth, which makes the hardware sampling system face great pressure; 2) in During the compression encoding process, a large number of coefficients with small amplitudes in the transform domain are discarded, resulting in a waste of data calculation and memory resources

[0011] Because most of the current compressed sensing reconstruction algorithms use fixed basis functions, that is, decompose the signal in a certain domain, such as: DCT domain, wavelet domain and gradient domain, but these domains ignore the non-stationary nature of the signal characteristics, lack of self-adaptive ability, so that the image cannot be decomposed sufficiently sparsely, which makes the effect of CS reconstruction very poor, and limits the application of CS in image

At the same time, in the learning process of sparse coding and dictionary, each block is considered independently, ignoring the correlation between blocks, which leads to inaccurate sparse coding

Images