Achievement

Perception in humans and machines requires rapid and accurate recognition of natural objects by means of shape cues. Trainee Wilder, working with Feldman and Singh, continued development of a stochastic model of natural shape recognition based on the idea that shapes are best represented in a skeletal format. Consistent with these ideas, the ability of human observers to detect contours and shapes in noise was found to depend on the model’s estimates of contour and shape complexity. This is surprising because detection only requires noticing a small segment of the contour; the entire shape does not need to be seen. The results support the idea that visual recognition operates best when using statistical rules to make decisions on the basis of global and holistic structures, rather than representations of arbitrary portions of the shape.