Segmentation and Recognition of Dimensioning Text from Engineering Drawings - Article - 1998

Oct 28, 2026

Engineering drawing understanding relies on accurate recognition of dimension text because it encodes the exact dimensions and tolerances of depicted objects. The work addresses dimension text processing in ANSI and ISO drafted drawings, starting with orthogonal zig-zag vectorization, arc segmentation, and arrowhead pair recognition. Textbox extraction uses region growing on text-wire candidates, followed by standard detection, logical textbox decomposition, neural network OCR on basic textboxes, and contextual verification by comparison with measurements directly on the drawing.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages6

Reading Time15 min

AccessFree to download

INTRODUCTION

BASIC CONCEPTS AND DEFINITIONS

Textbox and basic textbox

Logical textbox and subtextbox

COMPUTER VISION AND IMAGE UNDERSTANDING Vol. 69, No. 2, February, pp. 196–201, 1998 ARTICLE NO. IV960585 Segmentation and Recognition of Dimensioning Text from Engineering Drawings Dov Dori and Yelena Velkovitch Information Systems Engineering, Faculty of Industrial Engineering and Management, Technion, Israel Institute of Technology, Haifa 32000, Israel E-mail: [dori@ie.technion.ac.il](dori@ie.technion.ac.il), [elena@ie.technion.ac.il](elena@ie.technion.ac.il) Received July 25, 1995; accepted December 3, 1996 Recognition of dimensioning text in engineering drawings is an essential part of the drawing understanding process, as this text provides the exact dimensions and tolerances of the object described in the drawing. We consider engineering drawings produced according to either ISO or ANSI drafting standards. Text segmentation and recognition are preceded by orthogonal zig-zag vectorization, arc segmentation, and arrowhead pair recognition. Initial textbox extraction is done by a region growing process, performed on text-wire candidates. On the basis of textbox context (neighboring annotation wires) the drafting standard is detected. Raw textboxes are divided into logical textboxes, which are further decomposed into basic textboxes. A neural network based OCR algorithmisapplied to each basic textbox. Finally, theOCR recognition results areveriﬁed by using contextual information and comparing the results with the measurements made directly on the drawing. 1998 Academic Press Key Words: text segmentation; text-graphics separation; textgraphics association; engineering drawing understanding; CAD conversion; optical character recognition. 1. INTRODUCTION Currently, engineering and industry make extensive use of computer aided design (CAD) for creating and processing of engineering drawings. However, by now a great deal of paperbased engineering drawings have accumulated, and a few designers still prefer to use the conventional drafting techniques for producing the drawings. Since maintaining, modifying, and updating paper drawings manually is an expensive and timeconsuming process, it is desirable to have a system which will be able to understand paper-based engineering drawings and translate them into CAD representation [1] . An essential part of the drawing understanding process is recognition of dimension text. The problem of dimension text recognition may be roughly separated into the following subproblems: text and graphics separation, text string extraction, association of text with graphics, text string recognition and veriﬁcation. While several studies have been conducted which are related to these problems, we are not aware of any research that has been directly targeted at this goal. Fletcher and Kasturi [2] developed an algorithm for text string separation from mixed text/graphics images. It is based on the generation of connected components and the application of Hough transform to group together the components into logical character strings which may then be separated from the text. Lai and Kasturi [3] presented a system for detecting dimension sets in engineering drawings that are drawn according to ANSI drafting standard. It is also based on the generation of connected components and further composing them into text strings, which are associated with dimensioning lines. Neither one of these algorithms treats the problem of text/graphics connectivity nor do they detect text strings consisting of a single character. Chai and Dori [4] proposed an algorithm for textbox extraction that is preceded by orthogonal zig-zag vectorization, arc segmentation, and arrowhead recognition. The textbox extraction is done by clustering the remaining short bars that are close to each other through a region growing process. As noted by the authors themselves, the algorithm is designed only for detection of text areas without string extraction and may lead to some detection errors. This paper describes a s

FAQs

What role does dimension text recognition play in engineering drawing understanding?

Dimension text recognition is essential because it provides the exact dimensions and tolerances of the object described in the drawing.

How does the method extract and structure textboxes before OCR?

Textbox extraction is performed via region growing on text-wire candidates. Based on textbox context and detected drafting standard, raw textboxes are divided into logical textboxes, which are further decomposed into basic textboxes.

How are OCR results verified in this approach?

Neural network OCR is applied to each basic textbox, and recognition results are verified using contextual information and by comparing results with measurements taken directly on the drawing.

Oct 28, 2026

Upload by Nguyễn Văn Học