Dark light image adaptive enhancement method based on dense deep learning

Abstract

The invention discloses a dark light image adaptive enhancement method based on dense deep learning, and belongs to the field of computer vision. The dense deep convolutional neural network model provided by the invention comprises four sub-network models, so that local adaptive illumination enhancement, noise elimination, distortion repair and semantic repair of the dark image can be realized. The method comprises the following steps: firstly, decomposing a reflectivity diagram and an illumination diagram of the dark light image through a decomposition network; learning noise components with relatively small space in a reflectivity graph obtained through decomposition by combining illumination graph guidance information obtained through an MAMI module and a residual idea, and carrying out noise reduction; then, recovering semantic information of the reflectivity graph through a two-layer dense convolutional neural network and a loss function with a semantic recovery item, and restoring color distortion and the like; introducing the proportion information of the to-be-enhanced illuminance map and the target illuminance map and the gradient information of the to-be-enhanced illuminance map into the illuminance adjustment network, so that the global brightness can be flexibly adjusted and the local brightness can be adaptively enhanced; and finally, synthesizing the enhanced illumination image and the recovered reflectivity image to obtain a target image. According to the method, noise can be removed and distortion and semantics can be restored while the brightness of the image can be adaptively adjusted, so that the enhanced image is excellent in image aesthetic evaluation indexes and is superior to other methods in many computer vision tasks.

Description

technical field [0001] The invention relates to a dark-light image adaptive enhancement method based on intensive deep learning, which is used for semantic restoration and adaptive enhancement of images in dark-light environments, and realizes illumination enhancement, noise elimination, distortion repair, and semantic restoration of dark-light images , belonging to the field of computer vision. Background technique [0002] In recent years, with the rapid development of software and hardware such as the Internet and smart terminals, images have become a huge and important data resource. As we all know, image information plays a huge role in people's production and life. As an important information carrier, it promotes the exchange of information and helps people understand the world more intuitively. However, in real life, due to the limitation of natural conditions and technical conditions, there are a large number of dark light images in the massive image data. These lo...

AI Technical Summary

Problems solved by technology

Although these methods improve the quality of the image perceptually, they ignore the feeling of the "machine"

Images