Feature approach for printed document image analysis

Abstract

This paper presents advances in zone classification for printed document image analysis. It firstly introduces entropic heuristic for text separation problem. Then a brief recall on existing texture and geometric discriminant parameters proposed in a previous research is done. Several of them are chosen and modified to perform statistical pattern recognition. For each of these two aspects, experiments are done. A document image database with groundtruth is used. Available results are discussed.