Due to the severe light absorption and scattering, underwater images often exhibit problems such as low contrast, detail blurring, color attenuation, and low illumination. To address these issues, this paper presents a two-step strategy based on color restoration and image fusion by combining deep learning and conventional image enhancement technologies to improve the visual performance of underwater images. First, an adaptive color compensation method is proposed to make up for the loss of severely attenuated channels. Color restoration is further implemented to estimate the illuminant color cast caused by the selective attenuation of light. Since the underwater image after color restoration still suffers from scattering and blurring, an effective method based on dual image wavelet fusion (DIWF) and Generative Adversarial Network (GAN) is designed to further enhance the edge details and improve the contrast of the color restored image. Experiments demonstrate that the proposed method significantly outperforms several state-of-the-arts in both qualitative and quantitative qualities. The approach can achieve better performance of color restoration, blur removal, and low illumination enhancement.

由于严重的光吸收和散射，水下图像经常出现对比度低、细节模糊、颜色衰减和低照度等问题。针对这些问题，本文提出了一种基于色彩恢复和图像融合的两步策略，结合深度学习和传统图像增强技术，提高水下图像的视觉性能。首先，提出了一种自适应颜色补偿方法来弥补严重衰减通道的损失。进一步实施颜色恢复以估计由光的选择性衰减引起的光源偏色。由于色彩还原后的水下图像仍然存在散射和模糊的问题，设计了一种基于双图像小波融合（DIWF）和生成对抗网络（GAN）的有效方法，以进一步增强边缘细节并提高颜色恢复图像的对比度。实验表明，所提出的方法在定性和定量质量方面都显着优于几种最先进的方法。该方法可以实现更好的颜色恢复、模糊去除和低照度增强性能。