Color and texture count alike: An underwater image enhancement method via dual-attention fusion

Guodong Fan, Shuteng Hu, Jingchun Zhou, Min Gan, Chun-Lung Philip Chen

Abstract: Highlights•We propose DuAF, which models color and texture features in a targeted manner, fully leveraging their unique attributes and efficiently fusing key visual elements to achieve higher-performance visual information representation.•We introduce explicit semantic consistency constraints to enhance global sensitivity to pixel intensity distributions and color gradients, significantly improving color restoration in underwater images with complex color distortion. We design a local modeling method with dynamically adjustable perception windows, combined with relative positional bias optimization, to enhance the understanding of geometric relationships, enabling more efficient capture and reconstruction of intricate texture details.•We have designed an adaptive key feature fusion mechanism that optimizes the coordinated presentation of visual information through efficient collaborative Dual-Attention information.

External IDs:dblp:journals/inffus/FanHZGC26