Improving underwater visuals by fusion of Deep-Retinex and GAN for enhanced image quality in subaquatic environments

The enhancement of subaquatic images is crucial for various applications such as marine resource development, underwater photography, surveillance, and scientific imaging. However, the underwater environment presents challenges like color distortion, low contrast, and poor visibility, which traditional image processing techniques struggle to address effectively. In response, this study proposes an innovative approach named Deep‑Retinex‑GAN, which integrates Deep Retinex preprocessing and Generative Adversarial Networks (GANs) to refine underwater images. Initially, the subaquatic images are processed using Deep Retinex to separate them into reflectance and illumination components, reducing color distortion and enhancing contrast. Subsequently, the reflectance‑enhanced images are used as conditional inputs for a GAN model, allowing it to learn the mapping to a target domain with improved illumination, texture, and sharpness. Experimental evaluations conducted on both synthetic and real‑world underwater image datasets demonstrate the superior performance of the proposed method compared to existing techniques, achieving a PSNR of 34.741 dB, an SSIM of 0.978, and a CF(ΔE) of 8.2, as well as noticeable artifact reduction. Qualitative assessments further highlight the method’s ability to produce visually pleasing and realistic results. The proposed approach shows strong potential for a broad range of underwater applications, including photography, surveillance, exploration, and scientific research, by significantly enhancing the quality and interpretability of underwater imagery across diverse domains.