Super resolution reconstruction method based on total variation difference and convolution nerve network

Abstract

The invention provides a super resolution reconstruction method based on a total variation difference and a convolution nerve network. The method comprises an image decomposition step of decomposing an original low resolution image into a structure part and a texture part by using a method based on total variation difference, a structure part image amplification step of using a linear interpolation to amplify the structure part to obtain an initial amplification image, using a sharpening filter to sharpen an edge and finally carrying out result correction, a texture part image reconstruction step of using a linear interpolation to amplify the texture part, inputting an amplified image into the convolution nerve network, and carrying out operation to obtain a reconstructed texture image, and an image combination step of combining the amplified texture part image and the reconstructed texture image to generate a final super resolution image. According to the super resolution reconstruction image of the invention, the edge and structures of the image are maintained at the same time, the operation complexity is reduced, and the requirement of real-time performance is satisfied.

Description

technical field [0001] The invention belongs to the technical field of video image processing, in particular to a super-resolution reconstruction method based on total variable difference and convolutional neural network. Background technique [0002] With the popularization of digital products, images, as the main source of information for human beings, have been more and more widely used. At the same time, digital image processing technology has also been developed rapidly. The acquisition of video images is a key step in the digital image processing system. In the process of digital acquisition, the image resolution and image quality will be reduced due to the following factors: Sampling frequency - undersampling makes the frequency spectrum of the image aliased and degraded due to deformation effects; atmospheric disturbance, defocusing, The size of the sensor and the relative motion between the image acquisition device and the object to be photographed will cause imag...

AI Technical Summary

Problems solved by technology

[0008] Several existing reconstruction methods have the following defects: the method based on interpolation has low calculation, but the reconstruction effect is poor; the method based on reconstruction cannot reconstruct both the edge and the texture well at the same time; the method based on learning The method is computationally complex and has a strong dependence on the choice of training library

Images