Research On A Multi-Exposure And Color-Adaptive Low-Light Image Enhancement Algorithm

As shooting devices like mobile phones and cameras become more prevalent in our daily lives,images and videos have become increasingly significant in the everyday lives of individuals.Images and videos can not only record daily life,but are also widely used in various industrial scenarios,such as autonomous driving,object detection,etc.,Therefore,it is crucial to capture images that produce high-quality visual effects.However,various internal and external factors can impact the visual clarity of images and videos in real-life shooting scenarios,and among these factors is the lighting environment.In low-light environments,images and videos often look low brightness,loud noise,and blurry details,which not only fail to produce satisfactory visual experiences,but also fail to meet the needs of some key industrial scenarios.In low lighting,problems with image quality are inevitable.Although existing shooting equipment can improve image quality,expensive hardware is unaffordable for ordinary people.Therefore,the use of software algorithms to enhance low-light images is an alternative.This paper addresses this issue by exploring deep learning techniques,multi-level feature extraction mechanisms,and multi-exposure fusion theories.As a result,An approach to enhancing low-light images that is based on multi-exposure and color adaptation is proposed,including the following:In the opening chapter,the context and importance of research on enhancing low-light images are introduced,along with an overview of the current research status in this area.Chapter 2introduces some traditional low-light image enhancement methods,reviews the knowledge of deep learning and multi-exposure fusion theory,and finally introduces several image quality evaluation indicators used in this article.In the third chapter,a low-light image enhancement network is suggested,which relies on multi-level feature extraction and fusion techniques.The network consists of two sub-modules,feature extraction and feature enhancement,both of which make use of the attention mechanism of channel and spatial union to make the network pay more attention to the local area of underexposure and overexposure.The network first extracts image features at different scales and levels by the feature extraction module,and then these multi-level features are enhanced by the feature enhancement module,and finally the enhanced multi-level features are fused.The experimental outcomes demonstrate that the approach presented in this paper can effectively extract the feature information of the input image to achieve the purpose of enhancing the low-light image,but the result is color distortion.In Chapter 4,This paper suggests a method for enhancing low-light images that relies on multiple exposure and color adaptation.To deal with the brightness and color components of the image separately,the low-light image is initially converted from the RGB color space to the YCb Cr color space.Then,a convolutional neural network architecture consisting of two subnetworks is constructed,the first subnetwork uses the neural network proposed in Chapter 3 to enhance the Y channel.The second subnetwork is a color adaptive network,which adaptively adjustments on the C_b and C_r channel components.The final improved image is obtained by merging the results from both subnetworks through channels and converting them back into the RGB color space.By comparing the objective metrics and visual outcomes of the experimental results,it can be concluded that the approach suggested in this paper is successful in improving low-light images and produces pleasing visual results.