Vol. 56, Issue 2, pp. 171-202 (2026)

Abstract

Underwater image enhancement (UIE) is widely used in underwater photography and vision fields to improve the contrast and clarity of underwater images. Due to problems of light scattering from suspended particles, water absorption, and refraction, underwater images often suffer color distortion and low contrast issues. To address these issues, we propose a UIE method with a Transformer-based multi-scale spatial domain feature fusion network (TMDFNet). We design a multi-scale convolution extraction module to fuse multi-scale and multi-level information and enhance the network adaptability on image feature extraction across diverse scales and levels. Simultaneously, we develop a spatial domain feature fusion module to improve the perceptual quality of underwater images and optimize the parameter configuration of the network. Moreover, we create a transformer module to capture intricate spatial relationships in underwater images, which enhances the network’s comprehensive information extraction ability and the quality of reconstructed underwater images. A series of experiments are conducted to verify the effectiveness of TMDFNet. Quantitative and qualitative experiment results demonstrate that TMDFNet has competitive performance compared with other advanced UIE methods in subjective and objective evaluations.