This project will investigate the role of parts in the visual perception of objects. Roughly half of the cortex of the human brain is engaged in visual perception, and much of this cortex is devoted to constructing objects and their properties. The importance of objects extends beyond visual perception, because objects are central in human thought and language. Substantial evidence suggests that human vision represents many objects in terms of parts and structural descriptions. These parts are computed early in the stream of visual processing, and they dictate or influence the perception of figure and ground, the appearance of transparency, the judgment of shape similarity, the perception of symmetry and repetition, the classification and recognition of objects, and the learning of names for objects and their parts. This project will build on prior work to study several questions: How exactly does human vision divide objects into parts? How does it represent the qualitative and metrical structure of each part? How does it use parts to determine transparency, shape similarity, and object identity? Are different parts and part representations used for 2D silhouettes than for 3D objects? Are parts used for recognizing faces? How do parts influence language learning? Does object recognition employ viewpoint-dependent or viewpoint-invariant representations? The approach taken will be broad-based and multidisciplinary. Precise differential-geometric definitions of parts will be tested in psychophysical experiments with human observers. The psychophysical experiments will use both objective and subjective methods to assess the human perception of objects and parts. The resulting data will be used to create mathematical and information-processing models and to inform neural-network models. The information-processing models will be implemented in computer programs whose performance can then be compared against that of human observers, allowing further empirical testing of the models and refinement of the theories. This research will enhance understanding of a central aspect of human visual cognition. It will contribute valuable models and algorithms toward the construction of automated robotic vision systems that can see and recognize objects. It will constrain neural models and inform physiological experiments which explore the neural basis of biological vision.
