Abstract: Underwater images often suffer from color distortion and detail loss due to complex environments. Existing methods process color channels uniformly, ignoring their distinct characteristics, while Transformers underperform convolutional networks due to limited information utilization. To address these issues, a multi-scale attention and adaptive normalization-based underwater image enhancement network is proposed. The network consists of two stages: multi-scale feature extraction and feature enhancement with reconstruction. In the first stage, rich features are captured through multi-scale processing. In the second stage, an encoder-decoder structure is employed, incorporating a feature enhancement module and parallel skip connections to ensure feature integrity and structural consistency while maximizing information utilization. Experimental results demonstrate that the proposed network significantly improves color correction and detail preservation compared with existing methods, achieving superior qualitative and quantitative performance.

Key words: underwater image enhancement, deep learning, multi-scale, adaptive normalization, color space