Enhancement methodology for low light image

This is the authors' abstract. We don't add key points for this paper.

In order to solve the problems such as low brightness, high noise and poor contrast in weak illumination images, there are several methods proposed to address this issue.Usually, these methods are categorized into two different ways.One is based on traditional light-based technology, the other is based on machine learning technology.The low-light image enhancement is often a challenging task because the noises in dark areas are amplified with the overall brightness and contrast of the image.With the development of machine learning techniques, deep learning networks are becoming the popular research topics recently to overcome the disadvantages of noisy dots.Based on the deep analysis of the current research work, we proposed a novel network and carried out lots of comparison experiments to analysis the performances of these methods.By training, validation and testing on the datasets, the evaluation criterious are defined and utilized to analysis the efficiency of the methods.With the results, we draw the conclusion that the efficient low-light image method can make up for the shortcomings of the environment, bring better viewers' experience and provide preprocessing for subsequent high-level computer vision tasks, such as target recognition, face recognition, semantic segmentation, etc.