| ECVision indexed and annotated bibliography of cognitive computer vision publications This bibliography was created by Hilary Buxton and Benoit Gaillard, University of Sussex, as part of ECVision Specific Action 8-1 The complete text version of this BibTeX file is available here: ECVision_bibliography.bib
J. K. Tsotsos and S. M. Culhane and F. Cutzu
ABSTRACT
The modeling efforts described in this paper span theoretical considerations, computer simulations applied to real-world scenes and human psychophysics. The theoretical work initially addressed the question {"}Is there a computational justification for attentive selection?{"}. The obvious answer that has been described many times since at least Broadbent, namely that the brain is not large enough to process all the incoming stimuli, is not satisfactory since it is not quantitative and provides no constraints on what processing system might be sufficient. Methods from computational complexity were employed to formally prove for the first time that purely data-directed visual search in its most general form is an intractable problem in any realization (Tsotsos, 1989). There, it is claimed that visual search is ubiquitous in vision, and thus purely data-directed visual processing is also intractable in general. Those analyses provided important constraints on visual processing mechanisms and led to a specific (not necessarily unique or optimal) solution for visual perception. The constraints arose because vision was cast as a search problem; the combinatorics of search are too large at each stage of analysis and attentive selection is a powerful heuristic to limit search and make the overall problem tractable. Attention is an important mechanism at any level of processing where one finds a many-to-one convergence of neural inputs and thus potential stimulus interference, a conclusion reached in (Tsotsos, 1990). This was disputed by (Desimone, 1990); recently, however, experimental work seems supportive (for example, Kastner et al., 1998, Vanduffel et al. in press). The basic component of the attentional mechanism is a hierarchical neural network that implements a task-dependent, top-down, directed competition among conflicting neural elements, a circuit first described by Tsotsos (1993). As such, the mechanism implements a selective tuning of the visual processing hierarchy (the model is thus named the 'selective tuning model'). In contrast to theories that claim similar conceptual strategies for attention (such as Desimone and Duncan, 1995), our model has been fully detailed and simulated on a computer. It provides attentive control to a robotic camera system and attends both overtly and covertly to task-directed features and objects using real-world image sequences acquired from video cameras. As such, it is an existence proof that the key elements of the model are realizable and perform as expected. The exposition will proceed in two parts. The first part will overview the selective tuning model. In particular, the issue of attentional control will be highlighted. The second section will detail an experimental investigation that tests a basic prediction of the model, namely, that with attention, perception is impaired near the attended stimulus. |
|
| Site generated on Friday, 06 January 2006
|