Low Light Image Enhancement Algorithm Based on Improved Retinex

Most of the current low light image enhancement algorithms based on Retinex theory ignore the interaction between the reflection component and the illumination component, and some algorithms only take the reflection component as the enhancement target, which makes the enhanced image feel unnatural. In this paper, by optimizing the joint estimation of the reflection component and the illumination component, the solution model of the enhancement algorithm is transformed into a least squares problem and solved iteratively. In the iterative process, the estimated illumination component of each step is re-used in the estimation of the reflection component to fully release the interaction between the two components. In addition, we also propose an effective operator to extract fine details of the image to optimize the regular term, so that the enhanced image has richer details. Finally, we apply gamma correction to the estimated illumination component, making the image color more natural. On the three publicly available low light image datasets of LIME, MEF and NPE, the proposed algorithm and four mainstream algorithms in recent years are evaluated through three indicators that reflect the enhancement of image quality from different aspects. From the perspective of comprehensive performance of each indicator, the proposed algorithm has better comprehensive competitiveness.