Application of statistical pattern recognition to document segmentation and labelling.

摘要

In the field of computer analysis of document images, the problems of physical and logical layout analysis have been approached through a variety of heuristic, rule-based, and grammar-based techniques. In this paper we investigate the effectiveness of statistical pattern recognition algorithms for solving these two problems. Using a new software environment for manual page image segmentation and labelling, a dataset containing 932 page images from academic journals has been created. Several physical layout analysis algorithms have been implemented, including a new algorithm based on a logistic regression classifier. Three statistical classifiers were applied to the logical layout analysis problem, with encouraging results. A new model for how ink is laid out on a page was used to develop a prototype combined segmentation and labeling system. Finally, several applications have been investigated, and rudimentary implementations demonstrated. Results indicate that statistical pattern recognition approaches to these problems will be very fruitful.