VLSI architecture for computer vision based on neurobiological principles of organization
Biological and technological (wide-field-of-view) vision systems are confronted with the formidable task of managing data sets at a rate in excess of 10/sup 0/10 bits per second. In both cases, however, considering the required tasks, it appears that the data are highly redundant, and therefore must be reorganized before any type of higher level processing is applied to them. Reorganizations may include compression, and dimensional reduction according to the various relevant parameters. Biological processing at both the retinal and cortical levels often consists of repetitive simple operations applied to spatial and/or temporal neighborhoods, limited in their extend and duration. These are most adequate for the very high image data rates, in spite of the fact that those neurobiological systems actually consists of simple components which are several orders of magnitude slower than electronic components. It is the authors' goal to follow biological algorithms and principles of organization in the design of VLSI architectures, and to achieve similar or better performance in image processing and machine vision. Their efforts have yielded the following families of VLSI devices and systems. A highly parallel Intelligent Scan Image Acquisition VLSI sensing device has been constructed. It selectively scans only the relevant areas of interest in each image, thus effectively providing a compressed image for later processing stages. The device is controlled by an algorithm which is highly sensitive to image content. This sensor imitates the capability of the eye to concentrate on (attend) certain parts of the image, and even extends this by processing multiple focal points simultaneously. This is an example of how we applied the nonuniform sample-and-process algorithm, characteristic of biological vision, in a highly parallel architecture which surpasses the performance of the human eye.
- Research Organization:
- Dept. Technion - Israel Institute of Technology, Haifa (IL)
- OSTI ID:
- 6024074
- Report Number(s):
- CONF-8809132-
- Journal Information:
- Neural Networks; (United States), Vol. 1:1; Conference: 1. International Neural Network Society annual meeting, Boston, MA, USA, 6 Sep 1988
- Country of Publication:
- United States
- Language:
- English
Similar Records
VLSI design and synthesis for a class of two-dimensional problems
Neural-network algorithms and architectures for pattern classification
Related Subjects
42 ENGINEERING
36 MATERIALS SCIENCE
COMPUTER ARCHITECTURE
PARALLEL PROCESSING
OPTICAL SYSTEMS
VISION
ALGORITHMS
SILICON
MICROELECTRONIC CIRCUITS
ARTIFICIAL INTELLIGENCE
COMPUTERIZED SIMULATION
DATA PROCESSING
DESIGN
IMAGE PROCESSING
NEUROLOGY
PERFORMANCE
REAL TIME SYSTEMS
RETINA
SEMICONDUCTOR DEVICES
BODY
BODY AREAS
ELECTRONIC CIRCUITS
ELEMENTS
EYES
FACE
HEAD
MATHEMATICAL LOGIC
MEDICINE
ORGANS
PROCESSING
PROGRAMMING
SEMIMETALS
SENSE ORGANS
SIMULATION
990210* - Supercomputers- (1987-1989)
420800 - Engineering- Electronic Circuits & Devices- (-1989)
360603 - Materials- Properties