HBANet: Hybrid Boundary-Aware Attention Network

Luo, Xubo; Zhang, Jinshuo; Wang, Liping; Niu, Dongmei

HBANet: Hybrid Boundary-Aware Attention Network

Xubo Luo¹, Jinshuo Zhang², Liping Wang*³, Dongmei Niu²

¹Shanghai University of Finance and Ecnomics, ²Jinan University, ³Jinan Fourth Hospital
Computer Vision and Image Understanding (CVIU) 2024

Code

Abstract

HBANet unifies infrared (IR) and visible (VIS) imagery through a dual-branch encoder, a Hybrid Boundary-Aware Attention (HBA) module, and a lightweight decoder. The HBA module couples boundary-sensitive spatial attention with cross-domain feature exchange, enabling crisp edge preservation and faithful intensity reconstruction. Training leverages a hybrid fusion loss that balances structure fidelity, brightness consistency, and spatial smoothness.

BibTeX

@article{LUO2024104161,
	title   = {HBANet: A hybrid boundary-aware attention network for infrared and visible image fusion},
	journal = {Computer Vision and Image Understanding},
	volume  = {249},
	pages   = {104161},
	year    = {2024},
	doi     = {10.1016/j.cviu.2024.104161},
	author  = {Xubo Luo and Jinshuo Zhang and Liping Wang and Dongmei Niu}
}