Study On Low-light Image Enhancement Based On Convolutional Neural Network

Under low illumination conditions such as cloudy days,nights,and object occlusions,images obtained often have problems such as small dynamic range,severe loss of detail information,and a large amount of noise.Therefore,low illumination images can seriously affect or even limit the performance of the human eye or computer vision system.Enhancing the low-illumination image can enhance the sharpness of the image,highlight the texture details of the scene,and greatly improve the quality of the image,thus providing data quality assurance for tasks such as target recognition and tracking,and image segmentation.Therefore,studying the enhancement of low illumination images has important theoretical significance and practical application value.Focusing on the enhancement of low-illuminance images,the main work of this paper includes: firstly introducing the physical model of illumination imaging,analyzing the causes of low-illumination image formation,summarizing the characteristics of lowillumination images,and classifying low-illuminance images;secondly,simple The basic principle of convolutional neural network is introduced.Then,a low illumination image detection algorithm based on convolutional neural network is proposed.The network can automatically determine whether an image is a low illumination image.Finally,according to Retinex theory A low-illumination image enhancement algorithm based on convolutional neural network,which uses a full convolution network to decompose low-illumination images into illumination and reflectivity maps,then enhances the illumination map,refines the reflectance map,and finally According to the illumination imaging model,the enhanced illumination map and the refined reflectance map are synthesized to obtain a fusion result with better visual acuity.In this paper,the artificial data set is used for low-light image detection network training,and the synthetic data set and the real data set are used for testing,all of which have achieved high accuracy.The training of the low-light image enhancement network was performed using the synthesized data set,and similarly,the synthetic data set and the real data set were also tested.The test results show that the proposed algorithm not only can effectively improve the brightness of the image,but also improve the detail definition of the image to a certain extent,and can avoid color distortion and halo phenomenon to a certain extent.At the same time,the proposed algorithm has certain advantages in algorithm efficiency compared with existing algorithms.