Underwater Image Color Correction based on Surface Reflectance Statistics

Underwater image processing has attracted much interest during the past decades. Most of the underwater images suffer from the problems of backscattering and color distortion. In this paper, we focus on solving the problem of color distortion. Due to the light attenuation, which is caused by absorption and scattering, different colors of light will disappear gradually with the increase of water depth according to their wavelengths. The blue color has the shortest wavelength, so it can reach the largest depth, which results in the bluish tone of the underwater images. Our main contribution is that we proposed a new color correction scheme based on a local surface statistical prior. Our work mainly contains two steps. Firstly, we segment the underwater image into several non-overlapped blocks. Secondly, for each block, we estimate its illuminant based on the image formation model and the local surface statistical prior. By dividing the image block by its illuminant, the true reflectance can be obtained. Our experimental results demonstrate that our proposed method can achieve comparable or even better results than some state of the art approaches.