Research On Low-light Image Enhancement Methods

The ability of computers to process image information improved with the continuous development of information technology,and it has been used in many fields widely,such as face recognition,remote sensing image analysis,and automatic driving.Usually,the quality of the image itself directly determines the information processing result of the image.When it is a low-light image,due to its low brightness and poor quality,it not only makes it difficult for the human eye to observe,but also makes the machine vision system unable to accurately capture the image features,resulting in serious deviations and even wrong results.Therefore,it has important research value and significance in both theory and application to enhance the quality of low-light images effectively.In addition,the existing image evaluation indicators cannot objectively and accurately measure the brightness and color changes of the image in quality evaluation of the enhanced image.To this end,this paper mainly conducts related research based on lowlight image enhancement methods and image quality evaluation indicators.The main work is as follows:(1)In view of the problem that the current supervised deep learning enhancement algorithm for lowlight images mainly requires a large amount of high-low-light image matching data in practical applications,it is difficult to obtain corresponding data sets in practical applications,and a lot of computer resources are required in the model training process.This paper presents an unsupervised image cyclic enhancement network based on attention mechanism,which consists of a strong attention module and a cyclic enhancement sub-network.The former extracts important feature information of the image,and the latter enhances the overall brightness of the image through multiple cyclic iterations.The experimental results show that the model consumes less resources and runs fast.It can effectively improve the brightness of low-light images while retaining local texture information.(2)The low-light image deep learning enhancement algorithm cannot restore the image color effectively,a low-light image enhancement processing method based on homomorphic filtering and color correction is proposed.First,the low-frequency and high-frequency information of the low-light image is processed by homomorphic filtering,and then the color of the low-light image is restored by color cast detection and color correction.According to the characteristics of the]ow-light image,select the appropriate evaluation index to measure the brightness and color of the enhanced image.The experimental results show that the method can restore the color of low-light images.(3)According to the low-light image enhancement algorithm given in the above research,on the basis of in-depth analysis of image enhancement function requirements,a corresponding low-light image enhancement application system is designed,which mainly includes file module,image enhancement processing module,image quality evaluation module and the custom image enhancement module.It meets the actual demand of low light image enhancement algorithm based on this paperIn this paper,unsupervised image cyclic enhancement network based on attention mechanism and low light image enhancement processing method based on homomorphic filtering and color correction are proposed,and the selected objective evaluation indexes of image brightness and color provide a new solution for low light image enhancement and quality evaluation.The low light image enhancement application software system designed and developed provides technical support for its practical application...