The many forms of motion processing: how shape information affects motion recovery

Dominant models of visual processing commonly argue that shape and motion information generated by a moving object are separately processed in discrete, hierarchal neural pathways. There is now, however, a growing body of evidence showing that these two processes actually interact at a seemingly early stage of visual processing, contrary to the existing models. The central aim of the thesis is to examine the interaction of form information with three fundamental tasks of motion processing of increasing complexity; motion detection, direction processing and global motion integration. The thesis adds to this growing body of research and provides a novel model showing how and where in the visual system these interactions may occur.
The first study (Chapter 2) examined how form information alone can cause the perception of motion. This was done using the transformational apparent motion stimulus, where a shape change elicits the perception of motion in the direction explaining the change. The perceived motion direction is not predicted by motion energy models and instead shows that form information alone can cause motion to be detected. The study showed many transformational apparent motion stimuli can be globally integrated into a single, global percept, suggesting that the form information that causes the percept can enter the motion system by, at least, the stage of global motion pooling.
The second study (Chapter 3) examined an influential form-motion model that provides a potential mechanism for form information to influence motion processing. The model existing argues that the orientation cues generated by the extended integration time of V1 neurons (known as motion streaks) are multiplicatively combined in the same stage with a binary motion direction estimate. When this model was tested using visual aftereffects the orientation dependence of the after effect suggested that these form cues influence motion direction processing at a later stage of the processing hierarchy than predicted. A new model was developed that successfully accounted for the data where orientation-selective neurons in V1 modulate the gain of motion-selective neurons at a later stage of global motion integration.
The next two studies provide support for different aspects of the proposed model. A prediction from the model was first confirmed by showing that orientation information directly affects the stage of global motion integration (Chapter 4). Adapting to an oriented grating changed the perceived direction of a motion stimulus that was designed to null effects at the local motion level and with aftereffects instead most likely reflecting changes at the stage of global motion integration. The orientation and spatial frequency dependence of this aftereffect was also predicted by the model. The next study (Chapter 5) showed that the model predicts many instances where the orientation information associated with an object has been found to change its perceived direction. The result suggests that the shape of an object will modulate the gain of global motion selective neurons. These studies showed that orientation information most likely directly influences motion direction processing in the manner specified by the model.
The final study (Chapter 6) built upon the thesis’s key finding that form information enters the motion system by, at least, the stage of global motion integration. A well-known stimulus was used where an object translates behind opaque apertures, which has previously been claimed to show that form information can influence global motion integration. The study found little evidence that form information affected motion pooling. Instead, contrary to previous claims, integration was mainly determined by the representation of local motion information. The result did support a previous finding showing that shapes forming open contours are unlikely to be globally integrated, suggesting that some types of form information may act as segmentation.
Overall, the research presented in this thesis provides strong evidence that form information can influence motion processing and shows that this interaction likely occurs at the stage of global motion pooling. Furthermore a novel mechanism is provided that allows form information to directly influence motion processing.