Stroke feature extraction for lettrine indexing

Abstract

We commonly need to extract the image feature to understand and to realize the processes on it, and for each process we need a suitable feature extraction method. In this paper, we describe a stroke feature extraction method which will used in the lettrine indexing. A lettrine is a letter decorated who appear at the beginning of a chapter or a paragraph in ancient books. It's principally composed of strokes. This method is developed with the goal to characterize for indexing them by content using these particular components. We thus use strokes instead of pixels as elementary component. This study is innovative and first results are interesting. Indexing tests using this method will be made in the NaviDoMass project1.