Study On Image Enhancement Based On Retinex Theory

With the rapid rise of computing vision and artificial intelligence,image has gradually been one of the most important media for human to access outside information.It has been further studied in various fields as the foucus of current research,particularly in intelligent driving,drone shooting,medical clinical examination and remote sensing monitoring and the technology can be promoted in the related fields.However,due to the irresistible factors such as the irresistible factors such as the environment and light,image quality may be seriously degraded in the actual process of taking images.It is diffcult to get valuable information from the degraded images.Therefore,the enhanced technology is used to improve details and information of degraded images,even promote the development in its related fields,which has a strong industrial application value.In this thesis,two enhancement algorithms based on Retinex theory are proposed.To make the processed results meet the visual human eyes vision perception and solve the unnatural in Single Scale Retinex algorithm,the imbalance between overall vision and detail enhancement,an adaptive Retinex enhancement algorithm based on illumination compensation and detail adjustment is proposed in Single Scale Retinex framework.Firstly,based on the combination of weighted guided image filtering and morphology,a new image illumination component estimation method is proposed,which can effectively eliminate halo phenomenon and enhance details.The brightness adjustment factor and the detail adjustment model are added to the Single Scale Retinex model to solve the unnatural problems.A bee colony algorithm with the entropy is introduced to realize the parameter adaptation.To get optimal results in visually results,the combining the trigonometric function and the improved detail adjustment model,the illumination compensation and the detail adjustment of the original image are based on original image.In the fusion of compensation and adjustment results,some details may be lost and blurred,and a detail recovery scheme based on the adjustment between pixels is proposed.The final result is the enhanced results.The simulation results have shown that the proposed algorithm can achieve better results in both objective and subjective analysis.The above enhancement algorithm is calculated in the log domain,and the estimation method is used to solve the reflection of the image,which adds uncertainty to itself.To solve this problem,based on the variational framework,a variational Retinex enhancement algorithm based on illumination compensation and detail adjustment is proposed.Firstly,a new object function is constructed and the alternate strategy is used to solve the illumination and reflection of images.Illumination compensation and detail adjustment are directly performed on variational enhancement results.Illumination compensation is achieved by estimating the degree of darkness in images and combining trigonometric function principle.Details adjustment is realized by the normalization of function.Specially,in order to integrate the adjusted and compensated information effectively with the variational enhancement,the weighted values are designed from brightness,hue and exposure rate respectively.Finally,the final enhancement results are obtained by multiscale fusion.Public low-light datasets have shown the proposed algorithm is better than other algorithms from both objective and subjective analysis.