Characterising patterns of eye movements in natural images and visual scanning

Abstract

A method to characterise the pattern of human eye movements as a person views a scene is developed. Rather than describe the spatial distribution of fixations, a new concept is introduced–crossing number–that uses a single number to describe the complexity of the eye scan. First a gaze pattern, i.e. the curve formed by joining all the points of fixation in consecutive order, is generated. The number of times that the curve crosses over itself is the total crossing number. In two different experiments free eye movements were measured during viewing of natural scenes and the performance of a specific visual task requiring an eye scan through a randomly positioned set of items. We demonstrate that there are sometimes large individual differences in crossing number that may reflect different scanning strategies used by different observers.

Keywords:

• human
• oculomotor
• scanpath
• eye movements
• natural images
• crossing number
• knot theory

Acknowledgements

We thank John Findlay and Val Brown for giving us access to their data, forming part of experiment 2 and for comments on the manuscript. We thank Robert Martin and Wayne Smith for help with programming and data analysis. The project was supported by a discipline hopping award from the MRC and The Wellcome Trust.

Notes

Notes

1. The 10 specific images correspond to the following image codes from the van Hateren 27 database(s): 1261, 1306, 1430, 1474, 1482, 1483, 1486, 1498, 1683 and 3263.

2. If the reader wishes to compare performance here with that in the work of Findlay Brown 9, 16, our numbered observers correspond to their letters as: 1 = BR, 2 = JP, 3 = LW, 4 = LS, 5 = PB, 6 = SL.