Microsaccades uncover the orientation of covert attention

Engbert & Kliegl, Vision Research 43 (2003) 1035-1045

Our eyes perform small movements even while we look at stationary visual scene. These fixational eye movements are subdivided into tremor, drift, and microsaccades. Microsaccades are the fastest components of fixational eye movements. Here, we investigated the statistics of microsaccades in a classical spatial cuing paradigm (Posner, 1980). First, we reproduced microsaccade suppression with a minimum rate about 150 ms after cue onset - a well-known effect reported first by Findlay (1976), Winterson & Collewijn (1976), and Bridgeman & Palca (1980). Second, as a new finding we observe microsaccadic enhancement with a maximum rate about 350 ms after presentation of the cue. Third, we find a modulation of the orientation towards the cue direction. Therefore, attentional influences can bias the unconscious motor behavior during visual fixation. These results suggest that microsaccades can be exploited to map the orientation of visual attention in psychophysical experiments.

Animation: Fixational eye movements and microsaccades. In a simple fixation task rather erratic miniature eye movements are observed. The example represents a fixation with a duration of 2348 ms (or 588 data samples), recorded binocularly with an SMI EyeLink System (250 Hz). Microsaccades are small but rapid events which can be identified as approximately linear epoches of the eyes' trajectory (left eye: green, right eye: red). The original size of the fixation cross shown in the animation is 0.73°.  Data are the same as in Figure 1 of the above paper. (Created by Jochen Laubrock.)



Methods: Detection of microsaccades

Microsaccades are detected in 2D velocity space.  The detection thresholds are computed via a median-based estimation of the standard deviations of horizontal and vertical components of the eyes's velocity. We use a temporal overlap criterion to test for binocular microsaccades.  This procedure further reduces noise in the detection algorithm.  A MATLAB implementation of the algorithm with a short sequence of experimental data can be downloaded using the following files: