Training method and training model for image enhancement network, and image enhancement method

Abstract

The invention discloses a training method and a training model of an image enhancement network, and an image enhancement method. The training method of the image enhancement network comprises the following steps: acquiring a plurality of training image pairs; Inputting a first image in a training image pair to a pre-trained image generation network, outputting the image after a plurality of convolution processes, and calculating a first loss value of the output image relative to a corresponding second image in the training image pair according to the first loss function; Inputting an output image to a pre-trained image discrimination network, outputting a discrimination value after a plurality of convolution processes, and calculating a second loss value of the discrimination value relative to a preset tag value according to a second loss function; updating the network parameters of the image generation network and the image discrimination network in combination with the first and second loss values until the sum of the first loss value and the second loss value satisfies a predetermined condition, and the training is completed to obtain the trained image generation network as thetrained image enhancement network.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a training method and a training model of an image enhancement network, and an image enhancement method. Background technique [0002] With the development of Internet technology, people are increasingly dependent on quickly obtaining information, such as pictures and videos, through the Internet. However, a large number of pictures transmitted through the Internet have mediocre visual effects, and it is often difficult for Internet users to find pictures with good content and good image color. On the other hand, mobile terminals (such as mobile phones, tablet computers, etc.) have also become commonly used camera devices for people, but the photos taken by mobile terminals are difficult to meet higher visual requirements. Based on these two considerations, there are a wide range of application scenarios for improving the visual effect of images through image enhancemen...

AI Technical Summary

Problems solved by technology

However, this method has a single effect and is prone to problems such as unnatural adjustment effects and color blocks.

The development of Convolutional Neural Network (CNN, Convolutional Neural Network) has brought new ideas to image processing, and its enhancement effect is superior to traditional algorithms in some aspects. However, CNN-based algorithms are prone to problems such as unnatural transitions and color casts.

Images