USNET: UNDERWATER IMAGE SUPERPIXEL SEGMENTATION VIA MULTI-SCALE WATER-NET Provisionally Accepted
- 1Hainan University, China
Underwater images commonly suffer from a variety of quality degradation such as color casts, low contrast, blurring details, and limited visibility. Existing superpixel segmentation algorithms are challenging to achieve superior performance when they are directly applied to underwater images of quality degradation. In this paper, to alleviate the limitation of superpixel segmentation when applied to underwater scenes, we propose the first underwater superpixel segmentation network (USNet), which is specifically devised according to the intrinsic characteristics of underwater images. Considering the quality degradation, we propose a multi-scale water-net module (MWM) aimed at enhancing the quality of underwater images prior to superpixel segmentation. Then, for the scattering and absorption of light leading to reduce object visibility and blur edges, the degradation-aware attention (DA) mechanism is designed and integrated into MWM. This mechanism efficiently directs the network to prioritize regions with significant quality degradation, thereby improving the visibility of those specific areas. Both quantitative and qualitative evaluations demonstrate that our method can handle complicated underwater scenes and outperform existing state-of-the-art segmentation methods.
Keywords: Superpixel, Underwater Images, Image Enhancement, spatial information fusion, Neural Network
Received: 03 Apr 2024;
Accepted: 01 May 2024.
Copyright: © 2024 Wang, Duan, Luan, Liang, Shen and Li. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
* Correspondence: Mx. Hua Li, Hainan University, Haikou, China