Abstract
Object recognition is achieved through the integration of local features detected in early visual areas. We propose that the construction of surface in the early- to intermediate-level visual areas (V1-V2) contributes to the integration of local cues for the establishment of shape representation in higher visual areas. Physiological studies have shown that a majority of cells in V4 show the selectivity for the curvature and its position with respect to the object center as a primitive representation of shape (Pasupathy & Connor, 2001; 2002), and that the responses of the cells emerge gradually in two phases with the early response to contour fragments and the later response to curvature and its position (Yau et al., VSS2010). We investigated computationally the neural mechanism underlying the curvature selectivity, specifically, whether the responses are established through the integration of contour fragments based on the early representation of surface. We developed a model of V1-V2-V4 networks, and examined the behavior of the integration process and the emergence of the curvature selectivity. The analyses showed that the spatial pooling of V1-cell activities around the CRF center leads to the early phase of the V4 cells, and that the integration of the V1-cell activities based on the surface constructed in V2 corresponds to the late phase. The cellular preference for curvature and its position depended on the directional bias of the integration region with respect to surface. Furthermore, the model V4-cells showed the preference for acute curvature rather than obtuse curvature despite the fact that the model had uniformity in the directional bias, indicating a good agreement with physiology. These results support that the construction of surface in early to intermediate areas plays a crucial role in the integration of contour fragments into curvature for the representation of shape.
Meeting abstract presented at VSS 2012