상세정보

CONTENTS
List of Figures = ⅸ
List of Tables = xiii
Preface = xv
Acknowledgements = xviii
1 Introduction = 1
 1.1 Computational Complexities in Vision = 1
 1.2 Review of Multiprocessor Architectures = 4
  1.2.1. Mesh connected computers = 5
  1.2.2. Pyramid computers = 7
  1.2.3. Hypercube multiprocessors = 9
  1.2.4. Shared memory machines = 11
  1.2.5. Systolic arrays = 12
  1.2.6. Partitionable and hierarchical architectures = 14
 1.3 Organization = 17
2 Model of Computation = 19
 2.1 Parallelism in IVSs = 20
 2.2 Data Dependencies = 21
 2.3 Features and Capabilities of Parallel Architectures for IVSs = 25
 2.4 Examples of Integrated Vision Systems = 25
  2.4.1 Image understanding benchmark system = 25
  2.4.2 Motion estimation and object recognition = 26
3 Architecture of NETRA = 37
 3.1 Processor Clusters = 37
  3.1.1 Crossbar design = 40
  3.1.2 Scalability of crossbar = 40
 3.2 The DSP Hierarchy = 41
 3.3 Global Memory = 41
 3.4 Global interconnection = 43
  3.4.1 Interconnection network = 43
  3.4.2 Global bus = 43
 3.5 IVS Computation Requirements and NETRA = 44
 3.6 Comparison of NETRA with Other Architectures = 51
4 Parallel Algorithms on a Cluster = 55
 4.1 Classification of Common Vision Algorithms = 56
 4.2 Issues in Mapping an Algorithms = 57
 4.3 Performance Evaluation of Parallel Algorithms = 59
  4.3.1 2-D convolution = 60
  4.3.2 Separable convolution = 64
  4.3.3 Two-dimensional FFT = 65
  4.3.4 Hough transform = 69
 4.4 Parallel Implementation Results = 76
  4.4.1 2-D FFT = 78
  4.4.2 Separable convolution = 79
  4.4.3 Benchmark Algorithms = 80
 4.5 Summary = 82
5 Inter-Cluster Communication In NETRA = 83
 5.1 Alternatives for Inter-cluster Communication = 83
  5.1.1 Multistage interconnection network and global memory = 83
  5.1.2 DSP tree links = 84
  5.1.3 Global bus = 85
 5.2 Analysis of Inter-cluster Communication = 86
 5.3 Approach to Performance Evaluation = 90
 5.4 Performance of Parallel Algorithms on Multiple Clusters = 91
  5.4.1 Two-dimensional Fast Fourier Transform(2-D FFT) = 91
  5.4.2 2-D separable convolution = 102
  5.4.3 Hough transform = 109
 5.5 Summary = 116
6 Load Balancing and Scheduling Techniques = 119
 6.1 Need for Efficient Load Balancing Techniques = 119
 6.2 Load Balancing and Scheduling Techniques for Parallel Implementation = 120
  6.2.1 Uniform partitioning = 123
  6.2.2 Static scheduling(First-order scheduling) = 123
  6.2.3 Weighted static scheduling(second-order scheduling) = 125
  6.2.4 Dynamic = 126
 6.3 Parallel Implementation and Performance Evaluation = 128
  6.3.1 Feature extraction = 128
  6.3.2 Matching features = 131
  6.3.3 Time match = 137
  6.3.4 Second stereo match = 139
  6.3.5 Summary = 141
7 Concluding Remarks = 143
 7.1 Summary and Discussion = 143
 7.2 Extensions = 145
References = 147
Index = 153