To appear in Computational Neuroscience ‘96. (Ed J. Bower) Plenum Publishing, NY. Cortical Synchronization Mechanism for “Pop-Out” of Salient Image Contours Shih-Cheng Yen and Leif H. Finkel Department of Bioengineering and Institute of Neurological Sciences University of Pennsylvania Philadelphia, PA 19104, U. S. A. syen@jupiter.seas.upenn.edu leif@jupiter.seas.upenn.edu ABSTRACT We present a model based on long-range intra-cortical connections which computes the salience of contours in a visual scene. The model accounts for a number of psychophysical and physiological results on contour salience, and provides a mechanism for several of the Gestalt laws of perceptual organization. In the model, cells lying on smooth contours facilitate each other, and strongly facilitated cells enter a “bursting” model. Horizontal connections allow bursting cells to synchronize, and perceptual salience is defined by the level of synchronized activity. In particular, we propose that the intrinsic properties of synchronization account for the increased salience of smooth, closed contours 1. INTRODUCTION It has been suggested that cells in the supragranular layers of visual cortex with long-range horizontal connections might play a role in extracting salient features in a scene (Gilbert, 1992; Field et al., 1993; Kovács and Julesz, 1993). These cells have been shown to be sensitive to stimuli outside the classical receptive field, allowing contextual information to influence the response of the cell (Nelson and Frost, 1985; Kapadia et al., 1995). Similar cells in the supragranular layers of striate cortex have been observed to burst rapidly (Gray and McCormick, 1996) and could be involved in the temporal binding of contour elements. We present a model for computing the perceptual salience of contours that incorporates these two findings and is able to account for a number of physiological and psychophysical results (Polat and Sagi, 1993, 1994; Kapadia et al., 1995; Field et al., 1993; Kovács and Julesz, 1993, 1994; Kovács et al., 1996). 2. MODEL ARCHITECTURE Linear quadrature steerable filter pyramids (Freeman and Adelson, 1991) are used to model the response characteristics of cells in primary visual cortex. Steerable filters are computationally efficient as they allow the energy at any orientation and spatial frequency to be calculated from the responses of a set of basis filters. The fourth derivative of a Gaussian and its Hilbert transform were used as the filter kernels to approximate the shape of the receptive fields of simple cells. Model cells are interconnected by long-range horizontal connections in a pattern similar to the co-circular connectivity pattern of Parent and Zucker (1989), as well as the “association field” proposed by Field et al. (1993). For a cell of orientation θ A at location “A”, there is a “preferred” orientation at location “B”, φ B , given by the tangent to the unique circle which passes through both “A” and “B”, and whose tangent at “A” agrees with the local orientation, θ A , at “A”. The connection weights between the cell with orientation θ A , at “A” and the oriented cells at “B” peak at φ B , and decrease with increasing angular difference between the two orientations. These excitatory connections are confined to two regions, one flaring out along the axis of orientation of the cell (co-axial), and another confined to a narrow zone extending orthogonally to the axis of orientation (trans-axial). There is physiological and anatomical evidence consistent with the existence of both sets of connections (Rockland and Lund, 1983; Lund et al. , 1985; Nelson and Frost, 1985; Kapadia et al., 1995; Fitzpatrick, 1996). The connection field is shown in Figure 1a. As observed physiologically, these excitatory connections only facilitate cells that receive local supra-threshold input. If the local orientation activity distribution at “B” peaks at φ B , the cell with orientation θ A at “A” will be strongly facilitated. As the local orientation at “B” deviates from φ B , the degree of facilitation decreases. The “preferred” orientation at “B” can thus be thought of as providing “support” for the orientation, θ A , at