Block-processing adaptive filter for human eye movements

Abstract

Adaptive processing techniques can be divided into two categories: block processing and recursive methods. With block processing methods, incoming data are divided into blocks, and each block is processed as a whole to estimate predictor coefficients. With recursive methods, predictor parameters are updated as each new data point becomes available and computed thorugh a set of recursive algorithms. In this paper, five block processing adaptive filters are used in the prediction of the human eye movements. They are two-point-linear predictor (TPLP), five-point-quadratic predictor (FPQP), nine-point-cubic predictor (NPCP), polynomial-filter predictor 1 (PFP1), which is a linear convex combination of a TPLP and an FPQP, and polynomial-filter predictor 2 (PFP2), which is a linear convex combination of a TPLP, as FPQP, and an NPCP. These predictors were tested with various signals such as saccadic eye movements, sinusoidal, cubic, triangular, and parabolic signals. The results show that the TPLP is the best predictor for triangular signal and the NPCP is the best predictor for sinusoidal signal. Conversely, the FPQP is the best predictor for parabolic and cubic signals. The results also suggest that the PFP1 and PFP2 show significant improvement over that of the TPLP, FPQP, and NPCP in long-range prediction.