Vectorisation Of Engineering Drawings

Loading...
Thumbnail Image
Date
2010-01
Authors
Al-Kaffaf, Hasan S.Mohammed
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Document image analysis and recognition is an active area of research in the area of pattern recognition. It covers many topics such as recognition of on-line handwritten data, map interpretation, symbol recognition, and analysis and recognition of engineering drawings. Each of these topics consists of many steps such as digitisation, binarisation, segmentation, feature extraction, and recognition. This thesis is mainly concerned with image enhancement and feature extraction of mechanical engineering drawings. A procedure for enhancing salt and pepper noise removal from binary images of engineering drawings is proposed. The purpose of this procedure is to track and analyse thin graphical elements and to decide whether to retain or remove these elements. This procedure is integrated into three third party noise removal methods producing three proposed noise removal algorithms. Based on extensive experiment on mechanical engineering drawings corrupted by three noise levels (10%, 15%, and 20%) the three proposed noise removal algorithms have been shown to outperform six other noise removal algorithms in terms of Peak Signal-to-Noise Ratio (PSNR) and Distance-Reciprocal Distortion Measurement (DRDM) objectives measures. A new methodology for empirical performance evaluation of raster to vector conversion methods is also proposed. The methodology coupled with robust statistical analysis method can be used to study many factors that may affect the quality of line detection. Three independent factors are studied in this thesis namely noise removal, noise level, and vectorisation method. Experimental results have shown that all the three factors have a major role in affecting line detection. The results also have shown that performing the proposed noise removal algorithms prior to performing raster to vector methods leads to better line detection compared to other noise removal algorithms. A three-stage junction detection and resolution algorithm for thinning-based raster to vector conversion methods is also proposed. The algorithm works from low level representation (i.e. pixel representation) to high level representation {i.e. real junctions). Experimental results have shown that the proposed junction detection algorithm reduces the number of detected pseudo junctions and that performing the proposed noise removal algorithms prior to performing the proposed junction detection algorithms also reduces the number of detected pseudo junctions.
Description
Keywords
Computer Sciences , Engineering Drawings
Citation