MTNet: Multi-Task Underwater Image Enhancement Method Based on Retinex

Abstract

Underwater images play a critical role in underwater exploration and related tasks. However, due to light attenuation and other underwater factors, underwater images often suffer from color distortion and low contrast, which to some extent limit the efficiency and safety of underwater exploration. To meticulously address these issues and enhance the accuracy and reliability of underwater exploration, this paper proposes a multi-task underwater image enhancement method based on Retinex theory. This method divides the underwater image enhancement task into several sub-tasks, including image decomposition, color correction, detail reconstruction, and illumination adjustment. Specialized sub-networks— DecomNet, DecolorNet, and DelightNet—are designed to specifically address these problems, thereby alleviating color distortion, enhancing image details, and improving contrast. Experiments conducted on several publicly underwater image datasets indicate that the quality of underwater images is significantly improved after enhancement with the proposed method, compared to other representative underwater image processing techniques. For example, on the real-world dataset Underwater Image Enhancement Benchmark, the MSE, Structural Similarity Index Measure, and Peak signal-to-noise ratio scores achieved were 453.480, 0.901, and 25.145, respectively. This study holds significant implications for underwater exploration, with potential applications in the fields of marine research and underwater archaeology.OPEN ACCESS Received: 03/11/2024 Accepted: 27/12/2024 Published: 20/04/2025

Document