A survey of non-thinning based vectorization methods

Abstract

We survey the methods developed up to date for crude vectorization of document images. We classify them into six categories: thinning based, Hough Transform based, contour-based, run-graph based, mesh-pattern based, and sparse pixel based. The crude vectorization is a relatively mature subject in the Document Analysis and Recognition field, though there are rooms to improve. The purpose of the survey is to provide researchers with a comprehensive overview of this technique for them to choose a suitable method when developing their vectorization algorithms and systems. Keywords: Vectorization, Document Analysis and Recognition, Polygonalization