Depth color correlation-guided dark channel prior for underwater image enhancement

Due to the attenuation of underwater light, obtained images often suffer from low contrast, noise interference, and hazy content. In the field of underwater image enhancement, dark channel prior dehazing algorithms have achieved initial success, but they often do not take into account the correlation between different colors and depth. In this paper, we introduce the concept of depth-color correlation based on dark channel prior (DCP), additionally embedding a multi-scale dehazing and denoising module to effectively solve hazy content and noise interference. In particular, the depth-color correlation means that light from different color channels tends to have larger or smaller values as the depth increases. Therefore, we employ the three-bit indicator to represent the depth-color correlation and then further estimate the background light from the different color channels. Because the pyramid layers of the haze image contain varying degrees of noise, our multi-scale dehazing and denoising module decomposes the haze image into a Gaussian pyramid layer and a Laplacian pyramid layer, then dehazes and denoises them layer by layer. Thereafter, the pyramid layers are collapsed to obtain the enhanced image. Finally, in order to reduce the morphological artifacts, we employ the weighted guided filter. Results demonstrate that our method has superior enhancement capabilities compared to other state-of-the-art methods.