1.华中科技大学 人工智能与自动化学院;2.广州南方学院 大数据与人工智能专业
1.School of Artificial Intelligence and Automation,Huazhong University of Science and Technology;2.Big Data and Artificial Intelligence,Nanfang College Guangzhou
Underwater images suffer from the degradations such as color distortion and low-contrast, severely decreasing the image quality. The existing underwater image enhancement methods heavily rely on synthetic data and may face difficulties to generalize well in real scenes, due to the huge domain gap between synthetic and real images. To solve the problem, we formulate the challenging underwater image enhancement task into two easier yet physical meaning sub-problems: color correction and contrast enhancement. We propose a physically disentangled joint intra- and inter-domain adaptation paradigm, in which intra-domain adaptation focuses on color correction and inter-domain procedure transfers knowledge between synthetic and real domains. We first learn to physically disentangle haze images into three components complying with the scattering model: background, transmission map, and atmospheric light. Since color distortion is determined by scene light, we perform intra-domain adaptation by specifically translating scene light from varicolored space to unified color-balanced space, correcting the color distortion. Consequently, we perform inter-domain adaptation between the synthetic and real images by mutually exchanging the background and other two components. Extensive experiments demonstrate the superiority of the proposed method over the state-of-the-art for real varicolored water image enhancement.