imageprocessing.algorithms for stochastic image models sometimes are highly parallel and run slowly on conventional computers. An important question for pattern recognition and image database management is to find an...
详细信息
ISBN:
(纸本)0818606622
imageprocessing.algorithms for stochastic image models sometimes are highly parallel and run slowly on conventional computers. An important question for pattern recognition and image database management is to find an optimal computer architecture which fully supports the parallelism in these algorithms. For Markov image models, local neighborhood systems are essential to the probabilistic law. It is shown how to express imageprocessing.algorithms for Markov models using elementary operations of a neighborhood system so that a data-flow machine can execute these elementary operations in parallel.
An attempt is made to explain the particular characteristics of video processing. showing how they allow easy and efficient division into many parallel processes. Then the author gives more details of the transputer, ...
详细信息
ISBN:
(纸本)0903748592
An attempt is made to explain the particular characteristics of video processing. showing how they allow easy and efficient division into many parallel processes. Then the author gives more details of the transputer, before suggesting suitable architectures to use when connecting them in a multiprocessor array for video processing.
The authors analyze a particular parallel architecture that incorporates eight Texas Instruments TMS32010 digital signal processors. The fully built and operational multiprocessor system, named PT8, is described, and ...
详细信息
The authors analyze a particular parallel architecture that incorporates eight Texas Instruments TMS32010 digital signal processors. The fully built and operational multiprocessor system, named PT8, is described, and several common DSP algorithms are presented that illustrate its characteristics. It is shown that data-handling considerations are the major factors in the performance of these algorithms. Suggestions for upgrading this architecture to one that manages data more efficiently are given.
Optical implementations of digitalalgorithms that involve global operations on 2-D patterns are considered. Such applications of optical processing.will significantly speed up the response time of the pattern recogni...
详细信息
ISBN:
(纸本)0818606622
Optical implementations of digitalalgorithms that involve global operations on 2-D patterns are considered. Such applications of optical processing.will significantly speed up the response time of the pattern recognition systems, while the reliability offered by digital methods is maintained. The author summarizes some of these advances. He then proposes that a microcomputer-based hybrid system be developed with which digital implementations of algorithms involving neighborhood operations can be combined with optical implementations of algorithms involving global operations on 2-D patterns.
A reduced-mesh multiprocessor organization is proposed for parallel imageprocessing. Compared with fully shared-memory multiprocessor architecture, the reduced-mesh organization requires significantly less hardware a...
详细信息
ISBN:
(纸本)0818606622
A reduced-mesh multiprocessor organization is proposed for parallel imageprocessing. Compared with fully shared-memory multiprocessor architecture, the reduced-mesh organization requires significantly less hardware and can be implemented with a few VLSI chips. Many parallel computations, such as 2-D image transformations, FFT, least-square data fitting, and sorting can be realized efficiently on the reduced-mesh architecture. A linear speedup can be achieved on these computations with conflict-free accesses of the shared memories. This reduced architecture can simulate other cellular array image processors such as the mesh, the pyramid, and the hypercube. The reduced-mesh multiprocessor has better performance than other image processors in many applications. Modular expansion of the reduced-mesh architecture into a network of reduced meshes is proposed. This network extension further enhances the performance by supporting not only local but also global communications as required in FFT and sorting operations.
A highly parallel machine for imageprocessing.is presented. This machine is formed of identical processing.elements (PEs) interconnected according to a double structure: the PEs are arranged as orthogonal arrays to f...
详细信息
ISBN:
(纸本)0818606622
A highly parallel machine for imageprocessing.is presented. This machine is formed of identical processing.elements (PEs) interconnected according to a double structure: the PEs are arranged as orthogonal arrays to form different layers;these layers are connected, the PEs forming a binary tree. The architecture of the overall machine and of the individual cells is presented, and some aspects of its performance are discussed.
Following the development of several high speed video image processors, the lack of such special purpose processors for applications such as target tracking and imageprocessing.offering low power and volume was reali...
详细信息
Research is reported leading to the specification and design of low-cost microelectronic SIMD (Single-Instruction control of Multiple-Data streams) parallel-processing.modules for the cost-effective execution of a wid...
详细信息
ISBN:
(纸本)9024732026
Research is reported leading to the specification and design of low-cost microelectronic SIMD (Single-Instruction control of Multiple-Data streams) parallel-processing.modules for the cost-effective execution of a wide range of real-time 'imageprocessing.algorithms. The research project is dominated by the specification, design and fabrication of the SCAPE (Single-Chip Array processing.Element) chip, a 150K transistor CMOS chip which is fully optimised for the cost-effective implementation of such algorithms. Following a description of the algorithms, the author describes the chip architecture, chip floor plan, AMA allocation, operations, chain controller, and software of SCAPE. He concludes with a consideration of complexity and performance forecasts.
This issue contains 9 lecture course papers, 2 of which are in extended abstract form only. All of the papers have been abstracted and indexed separately. The topics covered include: parallel processing.reviews;parall...
详细信息
This issue contains 9 lecture course papers, 2 of which are in extended abstract form only. All of the papers have been abstracted and indexed separately. The topics covered include: parallel processing.reviews;parallel algorithms;Japanese research;multiprocessor systems comparison;parallel architecture;parallel languages and parallel processing.in artificial intelligence.
A digital optical processing.system consisting of optical gates and optical interconnections is described. Its concept has been demonstrated experimentally. The implementation of an optical cellular logic processor is...
详细信息
ISBN:
(纸本)0818606622
A digital optical processing.system consisting of optical gates and optical interconnections is described. Its concept has been demonstrated experimentally. The implementation of an optical cellular logic processor is considered. Optical systems for interconnections are given, and the architectural characteristics of such optical cellular logic machines are discussed and compared with electronic machines.
暂无评论