An improved method of Retinex for night color image enhancement

Images collected at night are usually characterized by low contrast, low brightness, missing numerous details and so on. In particular, an improved method of multi-scale Retinex is proposed for the case where there is a local light source, a large number of dark areas and high luminance regions in the image. Firstly, the adaptive brightness adjustment is adopted to replace the logarithmic transformation to compress dynamic range of the brightness components. Then, tri-gaussian model of the retinal neuron is introduced into the estimation of illumination image, and a bilateral filtering Retinex algorithm based on the tri-gaussian model and traditional Gauss filter is proposed. The image's edge detail can be better preserved while the contrast is effectively enhanced. At the same time, the color of the enhanced image can be maintained. Experimental results show that the algorithm is superior to other algorithms.