From Raster to Vectors: Extracting Visual Information from Line Drawings

Vectorization of raster line images is a relatively mature subject in the Document Analysis and Recognition field, but it is yet quite far from being perfect. We survey the methods and algorithms developed up to date for vectorization of document images and classify them into six categories: Hough Transform based, thinning-based, contour-based, run-graph-based, mesh-pattern-based, and sparse- pixel-based. The purpose of the survey is to provide researchers with a comprehensive overview of this technique to enable a judicious decision while selecting a vectorization algorithm for a system under development or a newly developed vectorization algorithm.