Research On Low-light Color Image Enhancement Algorithm

Enhancing low-light color images is an important research topic in the field of visual image processing and analysis.In reality,due to the limitations of imaging equipment and the influence of the imaging environment,the image quality is often unsatisfactory.Therefore,low-light image enhancement is a key content of the research of intelligent human-computer interaction technology,and it is also an important research direction in the field of computer vision.Because of its great theoretical research significance and wide range of practical application value,low-light-level image enhancement has received widespread attention and has achieved considerable development.Traditional low-light image enhancement is often aimed at simulating low-light images rather than low-light images in real natural scenes.And there will be insufficient enhancement or color distortion.Due to the efforts of academic community,algorithms for this problem have made great progress.However,there are still the following problems: 1)the enhanced images have color distortion;2)the enhanced image details are incomplete;3)the implementation of the image enhancement algorithm is difficult.To address the aforementioned problems,the main contents of this thesis can be summarized as follows.(1)A multi-scale low-light level image enhancement algorithm based on Retinex theory is proposed.This method extracts image features from multiple scales.Based on Retinex theory,it is composed of a decomposition part and an enhancement part.As a joint network,the decomposition and enhancement parts are mutually constrained,and the parameters are updated at the same time,so that the image processing results can be processed more excellent in detail,avoiding the problem of recombination after separation and enhancement,which will lose more information and increase errors.In addition,in the enhancement part,this thesis adopts a multi-scale network to fully extract image features,which can ensure that the illumination map achieves a balance between global and local brightness.Retinex theory can effectively solve the problems of noise amplification and color distortion,and at the same time add color loss to solve the problem of color distortion,so that the color of the enhancement result is closer to the normal light map.The enhanced results are excellent in visual experience,and the results of PSNR and SSIM also show the superiority of this algorithm compared to other algorithms.(2)A low-light image enhancement algorithm based on spatial pyramid is proposed.This method proposes a low-light enhancement network based on a spatial pyramid.The network first decomposes the image into two parts,a reflection map and a light map,and then enhances the light map.The enhanced light map and the reflection map form an enhanced image.When adjusting the light map,the spatial pyramid module is introduced,and more characteristic information of the image is obtained through three different-sized cavity convolution kernels.In addition,through the constraints of the color loss function,the result does not appear overexposed or blurred edges.Compared with the enhancement results of different algorithms,the algorithm performs well both quantitatively and qualitatively,and the algorithm has a significant effect on low-light image enhancement.(3)Realized the low-light image enhancement algorithm for image enhancement and optimized acceleration on the TX2 platform.The research transplants the low-light image enhancement algorithm from the computer platform to the NVIDIA Jetson TX2 platform,so that the algorithm proposed in this thesis can be implemented on the TX2 platform and the algorithm can be implemented on the embedded platform.In addition,this method also uses the Tensor RT inference optimizer to optimize the model of the algorithm proposed in this thesis,which solves the problem of low image processing efficiency,so that the algorithm of this thesis can play al role in actual engineering.