Abstract
Purpose: To determine whether subjects' gaze-fixation patterns (scanpaths) for representational paintings, viewed in a museum, are predicted by salient-feature models, a global-local viewing approach, content-information which forms the narrative, or aesthetic characteristics of the artwork. Methods: Four naïve observers viewed museum paintings, while wearing a battery-operated, headband-mounted ISCAN eyetracker. Two cameras simultaneously videotaped the eye and the scene; the data were digitized and analyzed off-line. Paintings varied in degree of representation, from French Neo-Classical to Impressionistic (e.g., Ingres to Monet). Afterward, viewers judged six aesthetic properties of each painting, using semantic-differential scales with endpoints labeled by adjective pairs, such as displeasing-pleasing, etc. Results and Conclusions: To quantitatively compare fixations with predictions based on Itti and Koch's (2000, Vis. Res.) saliency model, we conducted an analysis similar to Parkhurst, et al. (2002,Vis. Res.). Initial fixations did not support the model. Nor did scanpaths follow a global-to-local pattern. Rather, fixation patterns appear to be partly determined by narrative information; furthermore, scanpaths were rather idiosyncratic. Neither mean fixation duration nor viewing time was positively correlated with any aesthetic-judgment scores. The judgments for pleasingness were most often correlated with other scales. Results contradict current theories of scene perception and aesthetics that assume that areas of high saliency “drive” the eye.