Underwater Image Enhancement of Improved Retinex Base on Statistical Learning

Abstract

Underwater image enhancement technology is one of the key parts of underwater image processing, Since under water environment has greater attenuation and scattering effects on reflected light than the onshore environment that lead to uneven imaging, low contrast, color decay, etc. The underwater enhancement process is intended to highlight some key information of the image and eliminate the irrelevant information so as to make the preparations for subsequent scene segmentation, target recognition, target tracking and predictive analysis and processing. In this paper, we will focus on the problem of underwater image with low quality, and analyze the typical methods of box-based image enhancement methods. At the same time, Retinex image enhancement method based on optical analysis was introduced, and this paper proposed an improved enhancement method based on the Retinex methods. First we convert the image into HSV space for saliency analysis, next, learn the parameters of the harmonic HSV space by statistical learning. And then, we use the weights obtained by learning to predict the enhancement parameter for the decay image. Finally, the experiment verified that the method had better adaptability and better effect over the traditional Retinex algorithm in underwater image enhancement.