Single-frame image super-resolution method based on video coding

Abstract

The invention discloses a single-frame image super-resolution method based on video coding, and the method comprises the steps: carrying out the targeted processing of sub-blocks of different parts of an image through the prior information which can be directly obtained in the video coding, and processing the sub-blocks with more complex textures through a complex network; designing a self-adaptive convolution module to carry out targeted processing on sub-blocks of different coding modes, so the network is more targeted, different detail information is recovered for different textures, and the precision of a super-resolution result is improved. Parameters of a network with few channels are shared in a network with deep channels, that is, the super-resolution process of a whole picture is achieved through different layers of a backbone network, the relatively simple, shallow-layer and few-channel network is used for processing relatively large sub-blocks with smoother textures, and time needed by the super-resolution process is shortened.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a single-frame image super-resolution method based on video coding. Background technique [0002] Image super-resolution is the process of converting an input low-resolution visual image into a high-resolution visual image. An important focus of recent super-resolution work is to propose a variety of networks that accelerate the inference process. One of the branches is to achieve efficient super-resolution work with fewer parameters and faster speed. For example, the early FSRCNN directly extracts features from the input image, and then the feature map passes through an upsampling network to complete the construction of the super-resolution image. Another example is the recent work CARN, which uses group convolution technology to design a residual network to achieve fast processing of input images. Another branch is to increase the complexity of the network model, in...

AI Technical Summary

Problems solved by technology

[0004] In the traditional super-resolution network, the feature map is directly extracted from the whole picture. This structure makes the network unable to learn the different features of each region well, and the same convolution kernel is applied to process different regions to restore The texture details of the output image do not match the real image

And because the complexity of texture details in different areas of the image is different, the complex processing of low-detail areas often increases the computational load of the network unnecessarily.

However, as proposed by ClassSR, the neural network that classifies first and then passes through three parameters that are not shared will make it take a lot of time and computing power during training, increasing the complexity of the network.

In addition to the shortcomings mentioned above, most of today's super-resolution methods ignore the help of the prior information of the original image for the image super-resolution process.

Images