Research On Low Light Image Enhancement Algorithm Based On Deep Learning

At night or in low light conditions,due to the small number of photons collected and the low signal-to-noise ratio,the imaging device cannot accurately capture the details and colors of the image,which has a great impact on the quality of the image.Not only does this have an impact on human vision,but it also brings enormous challenges to computer vision tasks.Low-light image enhancement technology can reduce image noise and improve image quality,so it has great research value.In recent years,great progress has been made in lowlight image enhancement techniques,but existing algorithms still produce enhanced results with low contrast,overexposure,blurring and noise.In order to address this set of problems,this paper uses the theory related to deep learning to conduct research related to low-light image enhancement.The work in this article consists of the following sections:(1)In this paper,an LBP-based progressive feature aggregation network is designed to accomplish the low-light image enhancement task.LBP features are insensitive to illumination and contain rich texture information.In the network,LBP features are added to each iteration of the network in a guided manner,which helps to recover detailed information in low-light images.First,the global feature information in the original low-light image is extracted under a dual attention mechanism.Second,in the feature aggregation module,the different extracted features are aggregated.Next,a recurrent layer is used to share the features extracted at different stages and a residual layer is introduced to further extract deeper features.Finally,the enhanced image is output.The rationality of the method is verified by ablation experiments,and compared with many other state-of-the-art methods,the method has great advantages in both visual comparison and quantitative evaluation.(2)A multi-stage modular network model is designed for low-light image enhancement.The task is divided into three stages:a feature extraction stage,a feature aggregation stage and an image enhancement stage.Excluding the input and output layers,the network has a total of 88 layers,which complete the task of low-light image enhancement end-toend.The network extracts local and low-level features,such as colour and texture,from the bottom to the top layer.Finally,the global and local information are merged layer by layer.This method solves a series of problems existing in existing low-light image enhancement methods,and the effectiveness of the method is verified on multiple public datasets.The method is also compared with several existing methods for image enhancement.After extensive experimental verification,the algorithm proposed in this paper can effectively improve the quality of low-light images and outperforms state-of-the-art low-light enhancement methods in terms of several quantitative evaluation metrics.