Color paper map segmentation using eigenvector line-fitting

Abstract

This paper presents a color segmentation algorithm for paper-based United States geological survey (USGS) topographic maps. The algorithm uses an eigenvector line-fitting technique to overcome false colors introduced by the scanning process due to RGB misalignment. The RGB misalignment is caused by the optical characteristics of the scanner lens and is mostly evident in regions of color change. The resulting false colors render traditional color clustering schemes ineffective. The eigenvector line-fitting approach uses the color information of pixels in a 7 by 7 window to classify the color transitional pixels. This algorithm has been experimentally verified to be robust and accurate.