Share Email Print
cover

Proceedings Paper

Attentive sensing strategy for a multiwindow vision architecture
Author(s): Matthew J. Barth; Susan Hackwood; Gerardo Beni
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

A multi-window video architecture is a unique coarse-grained multiprocessor implementation used for image processing. Such an architecture consists of many local rectangular regions (windows) that are processed in parallel by individual processing units. The architecture provides many windows, whose position, size, shape and sampling rate are individually controllable. Each window captures image data into its own local memory, so image memory contention is eliminated. The motivation behind multi-windowing stems from the fact that there is often an excess of information in an image; we usually are only interested in a few objects that require further processing. By performing the processing in parallel only within windows, processing speeds are greatly increased. A key problem in multi-windowing is how to automatically assign windows to the relevant objects within an image. In order to address this problem, we introduce a strategy that models itself closely to the human attention ability, therefore we refer to it as an attentive sensing strategy. The attentive sensing strategy is carried out in two stages. The first 'pre-attentive' stage consists of several parallel processing units which rapidly extract salient information in parallel across the entire image. The second stage, termed 'attentive sensing', consists of a distribution of focused sensory processing on only the salient objects in the image. Information gathered by the pre-attentive stage guides the processing of the attentive stage, resulting in the elimination of irrelevant information. We carry out this sensing strategy on a multi-windowing vision system designed and built for the inspection of integrated circuit wafers.

Paper Details

Date Published: 1 July 1990
PDF: 14 pages
Proc. SPIE 1246, Parallel Architectures for Image Processing, (1 July 1990); doi: 10.1117/12.19580
Show Author Affiliations
Matthew J. Barth, Osaka Univ. (Japan)
Susan Hackwood, Univ. of California/Santa Barbara (United States)
Gerardo Beni, Univ. of California/Santa Barbara (United States)


Published in SPIE Proceedings Vol. 1246:
Parallel Architectures for Image Processing
Joydeep Ghosh; Colin G. Harrison, Editor(s)

© SPIE. Terms of Use
Back to Top