A study on the effects of noise level, cleaning method, and vectorization software on the quality of vector data

Abstract

Correct detection of line attributes by line detection algorithms is important and leads to good quality vectors. Line attributes includes: end points, width, line style, line shape, and center (for arcs). In this paper we study different factors that affect detected vector attributes. Noise level, cleaning method, and vectorization software are three factors that may influence the resulting vector data attributes. Real scanned images from GREC'03 and GREC'07 contests are used in the experiment. Three different levels of salt-and-pepper noise (5%, 10%, and 15%) are used. Noisy images are cleaned by six cleaning algorithms and then three different commercial raster to vector software are used to vectorize the cleaned images. Vector Recovery Index (VRI) is the performance evaluation criteria used in this study to judge the quality of the resulting vectors compared to their ground truth data. Statistical analysis on the VRI values shows that vectorization software has the biggest influence on the quality of the resulting vectors. © 2008 Springer-Verlag Berlin Heidelberg.