Following the erosion filtering, the slopes of potential
eyeblink waveforms are determined. Since an eyeblink is
characterized as a large negative slope followed by a positive
slope within a predefined time limit, a "low-noise"
Lanczos differentiation filter introduced in Hamming
(1989) is employed (see Figure 2c). The coefficients
ofan Nth order Lanczos differentiation filter are as follows: