Image compressed sensing reconstruction method based on residual dense threshold network

Abstract

The invention discloses an image compressed sensing reconstruction method based on a residual dense threshold network, and the method comprises the steps: carrying out the partitioning of an original natural image, carrying out the sampling through a Gaussian sampling matrix, and making a training set through the low-dimensional sampling amount and cut image blocks; recovering the low-dimensional sampling amount into a high-dimensional initial reconstruction image block through linear convolution; further optimizing and reconstructing the initial reconstructed image through the built residual dense threshold network; calculating a corresponding loss function, reducing loss through an optimizer, and carrying out back propagation on corresponding parameters; retaining the corresponding training model when the loss meets the requirement; and outputting the reconstructed image blocks through the training model and splicing the reconstructed image blocks into a final output image. Compared with an existing compressed sensing reconstruction method, the method has obvious advantages in reconstruction quality and reconstruction time.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to an image compression sensing reconstruction method based on a residual dense threshold network. Background technique [0002] Compressed sensing (Compressed Sensing, CS) reconstruction is a typical image technology, in the case of breaking through the theoretical requirements of Nyquist sampling law, by measuring the signal y∈R m×1 Recover the original signal x ∈ R as losslessly as possible n×1 (m<<n). Due to its wide application in the fields of image source coding and wireless broadcasting, compressive sensing reconstruction technology has received a lot of attention and research. The current compressive sensing reconstruction methods can usually be divided into two types: model-driven methods and methods based on Methods of Machine Learning Neural Networks. [0003] Among them, the model-driven method usually fully mines the prior information of the i...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the defects and insufficiencies of the existing technology, in order to solve the problems of poor interpretability, slow reconstruction speed and reconstruction quality in the existing reconstruction methods, the present invention provides an image compression sensing reconstruction based on residual dense threshold network method, which introduces the residual dense block with strong extraction features into the existing interpretable iterative threshold network framework, which can complete the reconstruction work with better detail quality and better interpretability

Images