This paper describes an improved algorithm for the automatic detection and correction of errors in scanned engineering drawings. The drawings are of simple machined components and have three orthographic projections with associated dimensioning and other textual information. The computer interpretation of scanned drawings is often made more difficult because lines in the original drawing are faded or very thin and are not detected by the scanner. We recently reported a method based on 3-D reconstruction, which can detect many cases of such missing lines but occasionally causes a line to be found incorrectly. This paper describes the evolution of our algorithm to avoid the detection of such phantom lines while still retaining the ability to detect lines which are genuinely missing.
Keywords: engineering drawing, error correction, 3-D reconstruction, CAD/CAM data, image restoration, image understanding.