Adaptive denoising method for EBAPS images captured under ultra-low-light conditions via scintillation noise...

Electron bombarded active pixel sensors (EBAPS), as vacuum-solid hybrid imaging devices, are promising for ultra-low-light imaging due to the high sensitivity and gain. However, image quality is severely degraded by complex mixed noise, particularly sparse and high-intensity scintillation noise resulting from ion feedback. To address this challenge, we propose a region-specific, edge-preserving denoising method tailored for EBAPS images, which suppresses noise while removing the impact of scintillation noise and preserving structural edges. We first construct statistical gray-level distribution models of scintillation noise under different bombardment voltages and develop a zero-mean normalized cross-correlation (ZNCC) based detection method to accurately locate and label scintillation noise pixels. Then, we introduce an adaptive filtering window size determination strategy driven by local gradient magnitude and directional standard deviation, enabling robust distinction between structural edges and flat regions in low-SNR EBAPS images. Two region-specific strategies are designed: linear directional sampling (LDS) for preserving structural edges in edge regions, and an adaptive side-window filter for noise suppression in flat regions. Experimental results demonstrate that the proposed method significantly reduces mixed noise while preserving structural details, achieving approximately 10% improvement in peak signal-to-noise ratio (PSNR) and 20% improvement in structural similarity index (SSIM) compared to other competing methods.