Image compressed sensing method based on self-adaptive nonlinear network and related product

Abstract

The invention mainly relates to the technical field of image processing. The invention provides an image compressed sensing method based on a self-adaptive nonlinear network. The method comprises thesteps: partitioning an original image to obtain at least one original image block, measuring the original image block through a preset sampling rate based on a convolutional neural measurement network, obtaining a measurement value of the convolutional neural measurement network; reconstructing the measurement value through a full connection layer; calculating to obtain an approximate solution ofthe original image block according to the measured value; learning a residual error between the original image block and an approximate solution of the original image block through an SRCNN network model; and obtaining a reconstructed image according to the residual error, enabling the calculation of image reconstruction to be simpler, shortening the time of image reconstruction, and further improving the quality of the reconstructed image through bilinear interpolation and extended convolution by utilizing the characteristic that the measured value extracted by the measurement network also retains the spatial information of the image.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to an image compression sensing method based on an adaptive nonlinear network. Background technique [0002] Compressed sensing, also known as Compressive sampling or Sparse sampling, is a technique for finding sparse solutions to underdetermined linear systems. In order to restore the analog signal without distortion, the traditional Nyquist sampling frequency should not be less than twice the highest frequency in the analog signal spectrum, and the large amount of data is not conducive to storage and transmission. In 2006, the compressed sensing theory proposed by Candes, Tao and Donoho can sample the signal at a frequency much lower than the Nyquist sampling frequency, and completely reconstruct the original signal with high probability. The main idea is to use the random measurement matrix Φ∈R m×n (m<<n) for the original signal x∈R n×1 Sampling (y=Φx,y∈R m×1 )...

AI Technical Summary

Problems solved by technology

In terms of reconstruction methods, convex relaxation methods, greedy matching pursuit methods and Bayesian methods are usually used to solve the corresponding sparse coding problems. However, real images cannot accurately satisfy the sparsity in the transform domain. The image quality reconstructed by the traditional reconstruction method is not high, which limits the development of image compression sensing technology, and reconstructs the image through multiple iterative optimization of a small number of measurement values, the calculation is complex and the reconstruction time is long, and it is difficult to achieve real-time performance.

Images