Symmetry Perception in Human Vision. Psycophysics and Modeling

Human brain functions well in dealing with visual information. When we look around, information provided by the surroundings is processed in such a way that a useful and rich representation of the world becomes perceptually available. Among other things, this process leads to the extraction of patterns. One basic characteristic of pattern perception is the distinction between figure and ground. Furthermore, we can perceive object features as shape, color, depth, orientation, location, and movement. We are also able to memorize visual forms, to recognize them, and to retrieve them after long time intervals. The study of how visual information is processed, stored, or recalled leads to the understanding of the structure and the function of our visual system. Mirror symmetry is an important property of natural and man-made objects. It can be perceived immediately and without effort. The perception of symmetry affects other visual processes such as the identification of a spatial reference frame, representation, memory, and recall. Mach (1886) was the first one who brought attention to the importance of symmetry in our understanding of the visual system. Since then, symmetry has been studied extensively and widely. Many phenomena have been observed. Hypotheses and models have been raised. Symmetry detection in random-dot textures is one way to study how patterns are processed and evaluated. This thesis is based on the studying of information processing in symmetry detection. Three groups of experiments were performed. The first group of experiments regards the influence of a property of the axis, its explicitness, on symmetry detection. Results shows that symmetry is more salient and easier to detect when the axis is explicit; which suggests that the identification of axis location facilitates the process of symmetry evaluation. The second group of experiments regards the relative importance of two aspects of vertical bilateral symmetry, the horizontality of virtual lines and the verticality of the axis. To study the relative contribution of these two components, two types of skewed symmetry were compared, following the paradigm recently developed by Johan Wagemans (1992). Results show that skewed symmetry with horizontal virtual lines and an oblique axis is easier to detect than skewed symmetry with a vertical axis and oblique visual lines, suggesting that the superiority of vertical bilateral symmetry over other types of symmetry is more explained by the horizontality of virtual lines than by the verticality of the axis. Matching along the horizontal dimension is easier than along oblique dimensions. The role of visual information about the 3D orientation of the pattern was studied too, since skewed symmetry corresponds to the projection of normal symmetry on a slanted or tilted plane. Results show that such information can improve the detection of skewed symmetry, especially when provided simultaneously. The third group of experiments explores symmetry detection in three-color patterns, to clarify the role of grouping processes involved in symmetry detection and the specific contribution of point-by-point matching. Two kinds of grouping were modified and compared: the grouping of symmetrical paired dots between two halves and the grouping of dots within each half. Same- and opposite- contrast patterns were used. Results show that symmetry is easier to detect when paired dots have the same contrast than when they have opposite contrast. This suggest that point-by-point comparison is the basic process involved in symmetry detection. Both kinds of grouping, grouping between and within, facilitates symmetry detection. Grouping between the two halves is more useful than within each half. The structure of the thesis is the following: Chapter 1 introduces theories of perceptual organization; Chapter 2 demonstrates phenomena observed in symmetry detection and presents some models accounting for the experimental results; Chapter 3, Chapter 4, and Chapter 5 describe details of Experiment Group 1, Experiment Group 2, and Experiment Group 3 respectively. Chapter 6 provides a model of symmetry detection with random-dot patterns, and presents the results of a computer simulation.