Research Group: Visual Attention and Perception

Modeling the Role of Visual Attention in Object Recognition Steffen Egner Issue. Recent models of human visual perception claim that visual attention plays a key-role in object recognition: To recognize a certain object in the visual field, the corresponding pattern is selected and then transformed into an object-centered reference frame. This representation is invariant to changes in size and position of the selected object. It therefore makes the subsequent recognition robust against differences in the size and position, in which an object appears in the visual field.

In GrKK since 10/94 as Ph.D. student

Goal: Present work aims to show, how visual attention may further enhance object recognition by selecting particular details of the previously selected object. This Zooming on a detail shall be described in a computational model, which then functions as a hypothesis for corresponding structures and processes in the human visual system.

Method: I extend an existing neural network model (Goebel, 1996) and provide it with a zooming-mechanism. The processes and structures in the model are (a) demonstrated by simulations, (b) compared to empirical data, and (c) analyzed according to other theories and models.

Problems: There is a lot of empirical data about object recognition and attention, but very little about the interaction of both.

Results: After focusing on an object in the visual field of the model, it detects whether the pattern can be classified unambiguously. If not, it uses information about details of the category in question, to (a) zoom into a particular detail, (b) generate an expectation about the visual properties of this detail and (c) compare the expectation to the actual properties of the detail. The model is in line with the empirical data and theoretical considerations.