Single-frame image super-resolution reconstruction method based on multiple differential consistency constraints and symmetric redundant network

Abstract

The invention relates to a single-frame image super-resolution reconstruction method based on multiple differential consistency constraints and a symmetric redundant network. The method comprises thefollowing steps: S1, establishing a consistency corresponding relationship between a low-resolution image and a high-resolution image, and respectively establishing observation models about a structure, an edge and a texture, that is, a consistency constraint model, by acquiring a 0-order gradient, a 1-order gradient and a 2-order gradient; S2, constructing a corresponding training set between thehigh-resolution image and the low-resolution image on the aspects of structure, edge and texture according to the type of the image to be reconstructed; and S3, establishing a training model based onthe symmetric redundant deep neural network, and obtaining a mapping relationship between the high-resolution image and the low-resolution image; and S4, establishing a super-resolution reconstruction model for the consistency constraint model in the step S1 and the prior constraint in the step S3 by using a semi-quadratic iteration method, and solving to obtain a reconstructed super-resolution image.

Description

technical field [0001] The invention relates to a single-frame image super-resolution reconstruction method based on multiple differential consistency constraints and a symmetric redundant network, and belongs to the field of computer image processing. Background technique [0002] Acquiring high-resolution images is an important foundation for computer vision and subsequent related fields. Current super-resolution reconstruction methods can be broadly scored into single-frame reconstruction algorithms and multi-frame reconstruction algorithms. The single-frame reconstruction algorithm refers to the algorithm required to reconstruct a corresponding high-resolution image from a low-resolution image affected by degradation factors such as noise, blur, and downsampling. For details, see literature W.Shi, J.Caballero, F.Huszar, J.Totz, A.P.Aitken, R.Bishop, D.Rueckert, Z.Wang, Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural N...

AI Technical Summary

Problems solved by technology

Algorithms that rely on the prior model are based on people's subjective will, and rely on previous experience to strengthen a certain feature while ignoring other features, which has a strong artificial tendency. For example: TV prior emphasizes edge protection and The protection of texture details is ignored, resulting in over-smoothing of the reconstructed image

The learning method relies on the external image library, which has two problems. 1. Whether the image library is complete, or some features cannot be recovered if it is not complete; 2. The internal structural information is not well grasped, and it mainly relies on external information for recovery.

Images