Extended Upscale and Downscale Representation with Cascade Arrangement

Smoothing filters are widely used in EEG signal processing for noise removal while preserving important features. Unlike common approaches in the time domain, a recent effective algorithm using the Upscale and Downscale Representation (UDR) technique has been introduced to process the signal in the image domain. The idea of UDR is to visualize the input with an appropriate line width, convert it to a binary image, and then smooth it by skeletonizing the signal object to a unit width and projecting it back to the time domain. We propose in this paper a cascaded UDR (CUDR) where the interested signal is filtered twice. CUDR’s performance is verified on simulated data with added white Gaussian noise and compared with the cascaded arrangement of some conventional techniques. Experimental results have demonstrated the outperformance of CUDR in terms of the fitting error when dealing with noisy signals, especially at a low signal-to-noise ratio.

I. Introduction

Electroencephalogram (EEG) is widely used for clinical diagnosis and monitoring to detect brain disorders [1], [2] as well as in Brain-Computer Interface (BCI) applications [3], [4]. One of common practices in EEG signal processing for specific pattern detection is via the visual inspection and interpretation of neurologists [5]. Therefore, the higher the EEG signal quality, the more promising results can be expected. Hence, EEG signal noise reduction becomes a vital aspect of EEG signal processing.