Wavelet-Guided Attention Network for Spiking Image Reconstruction

Long Chen, Hao Liu, Yuzhi Huang, Yaxin Li, Tingting Jiang, Jiangrong Shen, Qi Xu, Gang Pan

Keywords: Spiking Image Reconstruction, Wavelet-Guided Attention

Abstract: Reconstructing high-quality images from spike data remains a challenging task, particularly under low-light and high-motion conditions where spike noise and motion blur are prominent. To tackle these challenges, we propose WaveAttNet, a novel Wavelet-Guided Attention Network for spiking image reconstruction. WaveAttNet comprises two ky components: (1) a Wavelet-Guided Attention (WGA) module that performs frequency-aware noise suppression by emphasizing informative subbands and suppressing noisy ones through discrete wavelet transform and attention weighting; and (2) a Multi-scale Temporal Attention (MTA) module that captures and fuses temporal features across multiple time scales to mitigate both short-exposure noise and long-exposure motion blur. Extensive experiments on both synthetic (Spike-REDS) and real-world (Real-Captured) spike datasets demonstrate that WaveAttNet outperforms state-of-the-art approaches in terms of perceptual quality and quantitative metrics.

Primary Area: applications to neuroscience & cognitive science

Submission Number: 9909