Development of a non-linear smoothing filter for the processing of eye-movement signals.

The analysis of eye-movement (EM) signals poses problems for the designer of smoothing filters since many of the interesting types of EMs are bimodal. For example, optokinetic and/or vestibular stimulation results in an EM pattern called nystagmus consisting of alternating fast- and slow-phase components. Also, saccadic (refixation) EMs do not occur continuously, but are interspersed with periods of fixation. Conventional linear, low-pass filters (both finite impulse response (FIR) and infinite impulse response (IIR) types) smear the boundries between the fast- and slow-phases of nystagmus and the fixation and fast components of saccadic EMs. We have adapted a nonlinear smoothing filter (originally designed to optimize edge preservation in image processing applications) for the smoothing of EM signals. This filter is called a Predictive FIR-Median Hybrid (PFMH) filter. The PFMH filter operates on a moving window of data samples centered at the current point of interest. Several predictive FIR filters are applied to the "upper" and "lower" halves of the window and each are designed to predict the sample value at the center of the window. The median of these FIR filter outputs and the actual center data sample are taken as the PFMH filter output for each window position. By properly choosing the length and structure of the FIR subfilters, a PFMH filter can be designed to smooth a bimodal EM signal without blurring the boundries between the two signal components.