An efficient method for document image geometric layout analysis

Document image analysis is necessary for optical character recognition (OCR) and also very useful for many other document image manipulations. In this paper, we propose a document image geometric layout analysis system which has less region segmentation and classification error than that of the commercial software and previous works. The proposed method segments the document image into small regions to the size of a character using fast connected components generation method, so that it prevents the different types of connected components from combining. We also propose new criterion for clustering the connected components and some new techniques to deal with noise and reduce computation time. Experiment shows classification error rate of text and picture regions is decreased.