Night image enhancement method and system based on cyclic generative adversarial residual network and QTP loss item

Abstract

The invention provides a night image enhancement method based on a cyclic generative adversarial residual network and QTP loss items, and respectively improves problems faced by an unsupervised night image enhancement task through loss items of three dimensions of QTP. According to the invention, mixed loss comprises the loss items of the cyclic adversarial network, and three added parts, namely quality loss, task loss and perception loss, wherein the quality part solves a blurred image or false color problem by enhancing similarity between a reference image and an enhancement result quality score; the task part solves a problem of an insufficient enhancement effect from a perspective of constraining an enhancement result to have a higher daytime probability, namely, maximizing the daytime probability; and the perception part limits missing semantic information after domain transformation through a method of keeping Fourier phase spectrum of images before and after night enhancement to be consistent, and ensures content consistency of a night image and an enhanced image. In addition, a learnable and more ideal night image enhancement model is obtained by fusing new loss functions.

Description

technical field [0001] The present invention relates to the technical field of computer vision, deep learning and generative confrontational neural network, specifically a kind of dark night image enhancement method in the field of unsupervised real scene based on cyclic generative confrontational residual network and QTP loss item and its application in image processing Applications. Background technique [0002] With the rapid development of science and technology, computer vision technology is entering every aspect of people's life. For example, automatic driving, detection and recognition of scene cameras, auxiliary reversing images, etc. However, in the task examples mentioned above, the performance of computer vision technology is often better in an environment with sufficient light. Once it is applied in a poor light or even dark environment, the accuracy and performance of these visual tasks will drop significantly. The reason is that the visibility of the dark nig...

AI Technical Summary

Problems solved by technology

[0004] 1. False color, poor quality of image blur; 2. The enhancement effect of the night is not enough; 3. The distortion of the image before and after enhancement leads to the loss of semantic information

Images