Deep learning super-resolution method based on enhanced upsampling and discriminative fusion mechanism

Abstract

The invention provides a deep learning super-resolution method based on an enhanced upsampling and discrimination fusion mechanism, which comprises the following steps of: directly extracting original features of a low-resolution input image by using a single layer based on a deep learning convolutional layer in a residual error branch; using 6 cascaded multi-layer feature fusion-based circular convolution units to extract deep features, and enabling the deep features output by each circular convolution unit to be subjected to up-sampling through 6 deconvolution layers; fusing the up-sampled high-resolution features by applying a feature fusion mode with a discrimination mechanism, and performing dimensionality reduction on the up-sampled features by using a single-layer convolutional layer to obtain a residual error of the high-resolution image; carrying out up-sampling on the low-resolution image by using a bicubic interpolation method in a mapping branch; and adding the mapping and the residual of the high-resolution image pixel by pixel to obtain a final high-resolution image. According to the invention, an up-sampling opportunity is provided for low-resolution characteristics of each stage, and higher reconstruction accuracy is obtained.

Description

technical field [0001] The invention belongs to the technical field of low-level computer vision super-resolution, especially a deep learning super-resolution method based on an enhanced upsampling and discrimination fusion mechanism, which introduces a multi-stage upsampling and discrimination fusion mechanism, and self-adaptively calibrates and fuses The high-resolution features generated at each stage can achieve the purpose of improving the reconstruction quality, especially for large-scale reconstruction tasks. Background technique [0002] With the advancement of science and technology, more and more image resolution formats have appeared in people's daily life, from "Standard Definition" (Standard Definition) to "High Definition" (High Definition), and then to the current common " 1080p", "2K" or even "4K" and so on. Higher resolution means more image details, which in turn means that there may be a greater amount of information. Greater information volume implies g...

AI Technical Summary

Problems solved by technology

However, the upsampling module of these methods is too simple compared to the feature extraction module (only one layer of upsampling layer), which increases the instability and randomness of the network, and does not take into account the characteristics between the feature channels during feature fusion. The different importance of , thus achieving a more limited effect

Images