Synopsis
Discover how parallel geometric hashing handles scene data at scale, delivering fast recognition even with noisy points and thousands of models.
This book explains a parallel approach to model-based vision using geometric hashing. It shows how to precompute and index model features so recognition can be done efficiently on SIMD hardware. The discussion includes concrete algorithms, data structures, and implementation notes for parallel machines like the Connection Machine.
Readers will learn how to shape a recognition system that is independent of translation, rotation, and scale, while managing large model bases and noisy scenes. The text covers both offline preprocessing and online recognition, with practical details on histogramming, voting, and adapting hash tables for speed.
- How geometric hashing encodes models across many basis choices to enable reliable recognition.
- Techniques for fast voting and histogramming on parallel hardware, including radix-based methods.
- Strategies to distribute data and reduce communication when handling thousands of models.
- Trade-offs between preprocessing work and online recognition performance on SIMD architectures.
Ideal for readers of computer vision and high-performance parallel computing, especially those interested in scalable pattern recognition on parallel hardware.
"synopsis" may belong to another edition of this title.
