Automatic Diagrams Analysis

Abstract

This paper presents fully automatic approach analysis for information extraction from digitized grey level images of scanned diagrams in the field of graphics recognition. The proposed algorithms were tested on Telecom Egypt diagrams and some randomly selected diagrams. The analysis involves three distinct stages: the location of starting pixels in the diagrams; followed by a model based line-following to separate the text and drawings; then applying a fitting and vectorization algorithm on lines and circles in the extracted drawings. The information is feed into database system for later use by technicians' staff. The first step is to separate between the graphic components and the text associated to them. The segmented texts are recognized by OCR system. The diagrams are segmented into graphics components as lines, curves, circles and filled regions. The extracted information from the recognized texts is matched with the recognized graphic components and are feed into a database system. The algorithm extracts 95% of drawing lines (cables), and 100% of solid regions. The circle fitting and vectorization algorithm is capable of estimating 95% of the extracted corners, semicircles, and circles in the drawings, even the circles that are in a complex touch with the text inside them