An economical contour extraction algorithm for understanding large-size engineering drawings

An algorithm is described which produces the extraction of an object´s contours in a binary image while storing only two raster lines at a time during processing. The major data structure used is a block adjacency graph which occupies much less space (typically one hundredth) than the original image. Because this algorithm has very low costs of memory space and processing time, it is a powerful tool for processing large-size engineering drawings. Applications of the method to the recognition of some loop-structure logical symbols are described