Dark image enhancement method based on Retinex

Abstract

According to the dark image enhancement method based on Retinex, an image is converted from an RGB image to an HSV image, and the method is more in line with human vision. All the layers are directly connected together. In the novel architecture, the input of each layer consists of the feature mapping of all previous layers, and the output of each layer is transmitted to each subsequent layer. The feature maps are aggregated by a deep cascade. The depth of the neural network is improved, and the gradient disappearance problem is solved. An original network is up-sampled at the tail end of the network by using a nearest neighbor method to achieve a high-resolution size, compared with the original network, the improved EnhanceNet network adopts a bilinear interpolation up-sampling method, and by adopting the method, learning is not needed, the running speed is high, and the operation is simple. Only a fixed parameter value needs to be set, and the set parameter is a coefficient which needs to be multiplied by the central value. And compared with the original nearest neighbor algorithm up-sampling, the new up-sampling method can display the image more clearly, and the network effect is further improved.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a Retinex-based dark image enhancement method. Background technique [0002] With the advancement of science and technology, new image technologies are gradually being promoted, and people have higher and higher requirements for images in daily life. Images taken under low light conditions such as cloudy days or nights have low brightness, low contrast, and blurred details. Waiting for the problem, at this time you need to increase the brightness of the target a little. The traditional basic method is to process the frequency domain and the air domain. These basic methods are only aimed at image enhancement in specific scenes, and the effect of image enhancement in complex environments is not obvious. In recent years, image enhancement technology based on deep learning has made remarkable progress. It has achieved impressive results in advanced image understanding tasks...

AI Technical Summary

Problems solved by technology

Image enhancement technology based on deep learning algorithm has become a research hotspot. The conventional method is to directly train the dark image and bright image of the image data training set to obtain weight parameters. However, the visual effect and image quality need to be improved after the repair is completed.

Images