Multi images fusion Retinex for low light image enhancement

In this paper, a multiple-image fusion enhancement algorithm based on Retinex was proposed to solve the problem of contrast enhancement and naturalness preservation under low-light conditions. First, the maximum value was found in the R, G, and B channels to estimate the brightness of each pixel of the image as an initial illumination estimation. Based on the Retinex theory, the reflection image was generated and adjusted by morphological closing. Furthermore, the global contrast enhancement map and local natural degree keeping illumination map based on the initial illumination map were generated using a gamma transform and a double logarithmic transform, respectively. Subsequently, an adaptive weighted least square filtering fusion strategy was designed to fuse the three illumination images into the final illumination estimation image. Finally, the final illumination image was synthesized, and the reflection image was adjusted to obtain the image after the low-light enhancement. The experimental results indicate that the proposed algorithm has a lower lightness-order-error and natural image quality evaluator value compared to conventional enhancement algorithms. Moreover, the lightness-order-error and natural image quality evaluator values of real natural scenes can be reduced to 4.12 and 3.25, respectively, which yields better enhancement effects than conventional methods. Therefore, the proposed Retinex-based multiple-image algorithm using adaptive weighted least square filtering can effectively enhance the contrast and retain the natural degree of low-light images. © 2020, Science Press. All right reserved.