Udo Ernst, Universität Bremen
Information processing in the visual system is continuously challenged by the high-dimensional stream of sensory signals arriving from the outside world. For making sense of a visual scene, localized image patches have to be stitched together to form global representations, for example for recognizing objects. To facilitate this integration process, the visual system uses selective attention to enhance signals which are behaviorally relevant, while suppressing irrelevant information.
In my presentation I would like to discuss two aspects of visual processing in which decisions play a crucial role. First, I will focus on contour integration which is important for segmenting a visual scene. We studied this process by combining theory and modeling with psychophysical experiments. Human decisions in a contour detection task, in particular systematic perception ‘errors’, were used to constrain models, and turned out to provide the key for uncovering the neural mechanisms underlying this important computational process. In a second example, I will present results of a computational analysis of the temporal dynamics of neurons in area MT, and their modulation by attention. Here we identified a neural mechanism which quantitatively explains the large transient firing rate modulations that are typically observed in response to sudden stimulus changes. Most importantly, we could show that this mechanism allows attention to facilitate change detection independently on the sensory context, thus implementing an important principle of invariance for vision.