Seminar Series

Representing Dynamic Information for Object Recognition

We live and interact in a dynamic environment so we need to make sense of the changing visual input. Despite the change, there is structure in the pattern of change impinging on the retinas because of the inherent structure of the environment. For example, the retinal image of an object rotating rigidly in depth changes in a systematic manner because of the geometry of projection. Therefore, this structure-or dynamic information-can directly serve different visual functions. An increasing number of studies show that dynamic information plays an important role in perception and recognition. These studies reveal a hierarchy of visual processing stages. At the early stages, dynamic information is used for segmentation, shape recovery and diagnostic feature extraction. Importantly, these features are integrated at later stages for perception and recognition. Consistent with this hierarchy of processing stages, I present results which indicate that the visual system represents dynamic information by integrating shape and motion cues. In the first study, observers were required to detect dynamic targets in natural video sequences. In the second study, observers were tested with novel dynamic objects to control shape and motion cues available for recognition. Finally, a preliminary functional brain imaging study investigated the neural correlates of shape and motion integration for the recognition of novel objects. In light of these findings, different exploratory ideas of how the visual system may represent dynamic information at both the computational and neural level will be discussed.