Figure 3.

Blink detection algorithm. (A). Raw data (in pixels) from 40 trials of one participant. (B). Normalized data (trial by trial) forced to a mean of 300 (now in arbitrary units). Notice that in some cases, the natural fluctuations make using a fixed threshold (dashed horizontal line) for blink onset untrustworthy. (C). The benefits of non-linear trend removal, making the use of a fixed threshold for blink detection more robust. It can now detect partial blinks, yet not be triggered by large fluctuations, as these were removed. Colors simply indicate separate trials.