Introduction

Object identification involves the interpretation of two-dimensional (2D) images in terms of a 3D world. An image is nothing but a 2D array of color values in which the information from different objects is represented continuously without labels or explicit delimiters. A basic task of vision therefore is to detect the borders separating the regions corresponding to different objects (Fig. 1). These borders are called 'occluding contours' because they are generated by objects that partially occlude other objects in the scene. Occluding contours define the shape of the foreground object, but are unrelated to the background objects. Thus, the visual system not only needs to detect contrast borders, it must also assign 'border-ownership'.

The system uses binocular disparity as well as motion parallax and dynamic occlusion to recover the third dimension. Thus, binocular stereopsis and depth from motion are basic (and probably ancient) perceptual routines for identifying object structure in the visual sensory input. However, we can easily perceive objects and 3D layout of scenes also from pictures which provide neither stereoscopic nor motion cues. A white square surrounded by gray is perceived as a white figure on a gray background (Fig. 2a). Despite the absence of any depth information the visual system assumes a 3D layout. It interprets the square as an object, and the light-dark borders as the contours of the object. The Gestalt psychologists first showed that the visual system uses rules (Gestalt laws) to distinguish figure and ground. Fig. 2b demonstrates the compulsion of the system to interpret displays in 3D and to assign the contrast borders to one side or the other, as if they were contours of 3D objects (Rubin, 1921).

The underlying neural mechanisms are largely unknown. The Gestalt laws imply global visual organization, and border assignment has therefore been thought to occur at higher levels of the system such as IT cortex (e.g., Baylis and Driver, 2001). We have recently found that border-ownership is represented at stages as early as areas V1 and V2 (Zhou et al., 2000). Fig. 3 illustrates the influence of the location of a 'figure' on the responses of neurons in area V2. The cell of Fig. 3a responds more strongly to the edge of a light square above the receptive field than to the edge of a dark square below, although in both cases the receptive field (ellipse) is stimulated by exactly the same light-dark edge. In fact, the light- and dark-square displays are identical within the entire region occupied by the two figures, as shown by dashed lines on the right. Thus, the cell must have information about the global shape of the contours. Fig. 3b shows the ratios between the responses to preferred and non-preferred sides for 33 V2 cells. Nearly all cells showed the same side preference with the 3° and the 8° figure. Somehow, the global configuration modulates the local edge signals. We refer to this global influence as the Gestalt factor.

In this poster we report results on the influence of motion cues on neural border-ownership assignment. Motion parallax and dynamic occlusion are a powerful cues for border-ownership. When a region of stationary texture borders a region of moving texture, the moving texture is perceived as a surface extending behind the stationary texture and the border appears as the edge of the stationary surface (Fig. 4A). However, when the border moves together with the moving texture, the moving surface is perceived as in front, and the border appears as the edge of the moving surface (Fig. 4B). It was first assumed that the appearance and disappearance of texture elements at the border causes the depth stratification (Kaplan, 1969), but relative motion of contour and texture elements alone can also produce the stratification effect (Yonas et al., 1987).