Channel prior based Retinex model for underwater image enhancement

Since light undergoes absorption and scattering when it travels in water, images taken under water fall prey to color distortion, fuzziness/haziness, underexposure, and color cast. To alleviate these issues, this paper proposes a novel Retinex-based enhancing approach aimed at enhancing low-quality under-water images. Firstly, the input image undergoes pre-processing, where a color correction technique is employed to address the problem of color distortion. This is followed by conversion of the input image from the Red, Green and Blue color space to the Lab color space. Then, by means of the multi scale Retinex, the illumination component of the image is procured. Image dehazing algorithms, i.e., bright channel prior and underwater dark channel prior algorithms are applied individually on the procured illumination component. The image gets enhanced and is then converted back to the Red, Green and Blue color space from the Lab color space. The final enhanced image is obtained by performing histogram equalization on the enhanced Red, Green and Blue image. This is intended to make the output color intensity more realistic. The efficacy of the algorithms are gauged by means of some image quality metrics. Compared to pre-existing techniques, the method proposed manages to effectively enhance images while keeping detail loss to a minimum.