Please use this identifier to cite or link to this item:
|uhm_phd_9519432_r.pdf||Version for non-UH users. Copying/Printing is not permitted||4.93 MB||Adobe PDF||View/Open|
|uhm_phd_9519432_uh.pdf||Version for UH users||4.87 MB||Adobe PDF||View/Open|
|Title:||Automatic labeling, modeling and recognition for line-drawing interpretation|
Optical pattern recognition
|Abstract:||With the advent of the information era, line-drawing with digital form has become increasingly important in engineering applications. Unfortunately, there is a big media gap between paper and computer. For line-drawings to be useful and manageable they must be archived and understood by computers. However, the mere gathering of those digitized line-drawings certainly does not provide an economic way to store and retrieve. The need has always existed for a real-time system that automates the conversion process to obtain the symbolic descriptions of line-drawing images. Current technology for line-drawing interpretation involves excessive human supervision and is not easily extendible. Furthermore, it also fails to achieve both objectives of minimal model storage requirement and object matching time. A general paradigm for both 2D and 3D line-drawing interpretation systems is here developed and demonstrated with three major phases: labeling, modeling and recognition. The labeling module extracts a set of features known as symbolic labels of the corresponding objects from the image. These strategically selected labels facilitate automatic modeling and fast recognition dramatically. For interpreting 2D line-drawings, an automatic symbol segmentation approach for the electrical engineering drawings via the process of image blurring is first devised. A hierarchical neural network is then deployed for symbol modeling and recognition, thereby minimizing human intervention and achieving incremental extendibility capability. For interpreting 3D line-drawings. a linear-time-complexity polygon-division-based surface extraction algorithm for the projected trihedral objects is proposed. Then, a robust and efficient labeling approach is developed under a Cascaded Constrained Resource Planning (CCRP) model. Its near-linear-time complexity to the classical NPVI complete problems enables extensive usage of symbolic labels for modeling and recognition. Traditional viewer-centered object representation and matching approach requires excessive storage and computation time. Numerous less informative and redundant views are designated to be eliminated and thus gaining the efficiency for model searching. Each valid view is assigned a signature for automatic model base indexing. From the labeled line-drawing, valid-view modeling and multi-view matching are implemented to achieve the goals of lesser storage and faster retrieval time, which combine to realize a real-rime and geometrically invariant line-drawing interpretation system.|
|Description:||Thesis (Ph. D.)--University of Hawaii at Manoa, 1994.|
Includes bibliographical references (leaves 181-191).
xvi, 191 leaves, bound ill. 29 cm
|Rights:||All UHM dissertations and theses are protected by copyright. They may be viewed from this source for any purpose, but reproduction or distribution in any format is prohibited without written permission from the copyright owner.|
|Appears in Collections:||Ph.D. - Electrical Engineering|
Items in ScholarSpace are protected by copyright, with all rights reserved, unless otherwise indicated.