Abstract
Early vision algorithms typically can be classified as either pixel-based or image-based. Pixel-based algorithms such as edge detection and optical flow produce a dense output on every grid point. Image-level algorithms such as center of mass computation [1,5] produce a few outputs by integrating information from the entire image. Most special-purpose analog VLSI vision chips developed so far implement early vision algorithms of these categories. Little work has been done in building special-purpose chips that move beyond these early-vision algorithms. Dynamic wires, first introduced by Liu and Harris [2], provide an avenue for object-based processing. The dynamic wire model is a methodology that provides dedicated lines of communication among groups of pixels that share a common property. The methodology consists of first, configuring the switches in a 2D network for the groups of pixels that share a common property and second, utilizing the resulting dynamic connections for computation. These tasks can be accomplished by using a network of resistors and switches (Fig. 1). Examples of features which may be computed from the output of an early vision module like edge detection are length of a contour and the area inside a contour.
Recommended Citation
C. Koch et al., "Object-based Analog VLSI Vision Circuits," IEEE Intelligent Vehicles Symposium Proceedings, pp. 74 - 78, article no. 252236, Institute of Electrical and Electronics Engineers, Jan 1992.
The definitive version is available at https://doi.org/10.1109/IVS.1992.252236
Department(s)
Electrical and Computer Engineering
Document Type
Article - Conference proceedings
Document Version
Citation
File Type
text
Language(s)
English
Rights
© 2025 Institute of Electrical and Electronics Engineers, All rights reserved.
Publication Date
01 Jan 1992
Comments
National Science Foundation, Grant None