Parallel architectures for vision
- Dist-Univ. of Genova, 16145 Genova (IT)
- Thomas J. Watson Research Center, IBM, Yorktown Heights, NY (US)
- IBM Almaden Research Center, San Jose, CA (US)
Vision computing involves the execution of a large number of operations on large sets of structured data. Sequential computers cannot achieve the speed required by most of the current applications and therefore parallel architectural solutions have to be explored. In this paper the authors examine the options that drive the design of a vision oriented computer, starting with the analysis of the basic vision computation and communication requirements. They briefly review the classical taxonomy for parallel computers, based on the multiplicity of the instruction and data stream, and apply a recently proposed criterion, the degree of autonomy of each processor, to further classify fine-grain SIMD massively parallel computers. They identify three types of processor autonomy, namely operation autonomy, addressing autonomy, and connection autonomy. For each type they give the basic definitions and show some examples. They focus on the concept of connection autonomy, which they believe is a key point in the development of massively parallel architectures for vision. They show two examples of parallel computers featuring different types of connection autonomy - the Connection Machine and the Polymorphic-Torus - and compare their cost and benefit.
- OSTI ID:
- 6525071
- Journal Information:
- Proc. IEEE; (United States), Journal Name: Proc. IEEE; (United States) Vol. 76:8; ISSN IEEPA
- Country of Publication:
- United States
- Language:
- English
Similar Records
Parallel architectures and parallel algorithms for integrated vision systems
Parallel algorithms and architectures for binary image component labeling