Single-trial ERPs denoising via collaborative filtering on ERPs images

Abstract

The single-trial denoising is still a challenge in both the neuroscience research and signal processing due to poor signal-to-noise ratio. The Event-related Potentials (ERPs) denoising using ERPs image processing has received a significant attention in recent years. Considering the importance of latency and amplitude details in the ERPs analysis, the desirable methods for ERPs denoising are supposed to remove the large noise effectively while keeping the important information of the ERPs signal, latency and amplitude. A collaborative filtering that includes image patch grouping, shrinkage in 3D transform domain, and aggregation is applied to remove the background noise from the ERPs images. The denoising experiments have been evaluated on simulated data and real data using waveform observation, objective criteria calculation, and single-trial classification. The validations have demonstrated that the collaborative filtering is able to remove the noise effectively compared to wavelet and non local means. Moreover, it also preserves the details of the ERPs signals for latency and amplitude estimation simultaneously.