On a Parallel Implementation of Geometric Hashing on the Connection Machine (Classic Reprint): Technical Report 554: Technical Report 554 (Classic Reprint) - Softcover

Rigoutsos, Isidore

Softcover

ISBN 10: 1332092268 ISBN 13: 9781332092260

Publisher: Forgotten Books, 2018

Synopsis

Explore how geometric hashing speeds object recognition on parallel hardware.

In this study, the authors describe a parallel implementation of geometric hashing on the Connection Machine. The work covers how a model database is prepared off-line and how scene features are matched quickly in real time. It highlights the balance between computation and storage, showing how parallelism reduces the number of tasks while increasing memory use. The report also discusses how the approach handles rigid and similarity transforms and what happens when noise appears in the input.

- How the hash-based approach distributes work across many processors for fast recognition
- The role of rehashing, foldings, and symmetries in saving time and memory
- Performance results under different data distributions, scene sizes, and noise levels
- Practical insights for implementing parallel vision algorithms on large, multi-processor systems

Ideal for readers of high‑performance computing, parallel algorithms, and model-based vision who want a concrete look at scalable object recognition techniques.

JSON_OBJECT_PLACEHOLDER

About this title

Product Description

Excerpt from On a Parallel Implementation of Geometric Hashing on the Connection Machine: Technical Report 554 We report on a scalable parallel implementation of geometric hashing on a Connection Machine. The algorithm that is employed has been described in [9]. Using the resulting implementation, it is possible to recognize models consisting of patterns of points embedded in scenes, independent of translation, rotation, and scale changes, when there are thousands of models containing approximately 16 points each, with scenes consisting of hundreds of points, where most of the scene points are spurious noise points, and where embedded model points in the scene may be obscured or misplaced. With 1024 models and a scene of 200 points, the implementation yields an execution time of 70 milliseconds per probe on a 64K-processor Cm-2 parallel computer. Most of the execution time is taken in performing histograms of (model, basis-set) records. The algorithm is scalable, yielding an expected execution time that is O(log M + log S log M) in a Mn -processor hypercube-connected SIMD machine such as the Connection Machine. M is the number of models, n is the number of points per model, and S is the number of scene points. We also describe and report on a series of experiments for both the similarity and rigid transformation cases; these experiments provide information about detection and false alarm rates for varying amounts of noise in the input. About the Publisher Forgotten Books publishes hundreds of thousands of rare and classic books. Find more at www.forgottenbooks.com This book is a reproduction of an important historical work. Forgotten Books uses state-of-the-art technology to digitally reconstruct the work, preserving the original format whilst repairing imperfections present in the aged copy. In rare cases, an imperfection in the original, such as a blemish or missing page, may be replicated in our edition. We do, however, repair the vast majority of imperfections s

Edition details

• Publisher: Forgotten Books
• Publication date: 2018
• Language: English
• ISBN 10: 1332092268
• ISBN 13: 9781332092260
• Binding: Paperback
• Number of pages: 31