THINKING THROUGH DRAWING: PRACTICE INTO KNOWLEDGE

(Jeff_L) #1

76 TEACHERs COLLEGE COLUmbIA UNIvERsITy


visuomotor Atoms of Copy-Drawing


stages. The snapshots correspond to the follow-
ing observed sequence: hand stops—fixation(s) on
the left—saccade—fixation(s) on the right—hand
moves. We interpret the points where the hand
stops as key points, at which the hand’s action
needed to be reprogrammed and thus fixations on
the original image became necessary. The general
tendency of the gaze to move orderly along the
image contour (Figure 2) and to do so in parallel
with hand movements, suggests that this peculiar
form of the scanpaths is a precise eye-hand coordi-
nation strategy in support of graphical continuity of
drawing gestures. See Figure 3.


A computational model of visuomotor coordina-
tion reproduced the copy-drawing scanpaths sig-
nificantly better than image features alone; but both
performed not better than chance under free-viewing.


The scanpaths discussed above were the result, we
argued, of a dynamical coupling between eye and
hand movements. To further specify this intuition,
we introduced a computational model of eye move-
ments that combined two main sources of informa-
tion: 1) low-level image features, analogous to those
processed in early levels of the visual system (e.g.
local brightness, contrast, orientation), were com-
bined according to an established model of visual
salience (Itti & Koch 2001); 2) association prob-
abilities between planned eye and hand movement
directions, in response to the visual input, were
learned by the model via supervised training on the
drawing task. Further model details were provided
in (Coen-Cagli et al. 2008).
We tested variants of the model on the same
images used in the experiments, and quantified the
similarity between the resulting sequences of eye

Figure 2. a) Average distance between pairs of fixation points as a function of the number of saccades, in the
two conditions, across all subjects and trials. Distance is expressed in pixels, with 1 pixel corresponding to
ca. 0:05deg. Error bars denote 95% confidence interval. b) Example scanpaths superimposed on the original
image; red circles denote fixation points, the black circles denote the first fixation.


Figure 3. The sequence of eye and hand
movements by one subject in the draw-
ing task. In the upper row, cumulative
fixations on the original image are rep-
resented by red circles. In the lower row
the solid black square denotes the gaze
point; in the rightmost panel, the black
circles denote the endpoints of each
trajectory segment.

Free download pdf