Abstract: In view of the serious degradation of the underwater images caused by underwater light
attenuation and scattering, a new method of underwater image enhancement is proposed in this paper.
The new method based on the principle of dark channel prior conducts underwater image
enhancement and image thinning on dark channel by the method of soft matting. In the first 0.1%
brightest pixels of the image, the medians of these pixels are calculated through the median filtering
algorithm. Then, the differences between these pixels and their corresponding medians are calculated,
and the pixel of minimum difference is employed as an estimate of the background light. The
accuracy of the estimated background light is judged by the color saturation variance of the region
around the pixel. Afterwards, the Retinex algorithm and the ratio of attenuation coefficients for each
color channel are adopted to conduct color correcting for enhancement images. It is shown by
experiments that the proposed method can effectively dehaze the underwater images, correct the color
cast of images and further enhance the image contrast.

Key words: underwater image enhancement, dark channel prior, background light estimation, color
correction