Structure analysis of low resolution fax cover pages

Abstract

The increase in the use of faxed documents calls for the need to handle them automatically and intelligently for efficient storage, retrieval and interpretation. A lot of work has been accomplished for page segmentation in high resolution document images. But conventional methods for page segmentation are not suitable for faxed document processing. The well-known difficulties in faxed document processing are concerned with low resolution images and non-standardized formats. In this paper, we propose an effective structure analysis method for low resolution fax cover pages, based on region segmentation and keyword recognition. The main advantages of the proposed method are its capability of accommodating various types of fax cover pages and its fast processing speed. We divide fax cover pages into three regions - header, sender/recipient information and message - to easily identify the recipient's field. The recipient's name is then extracted through the recognition of keyword. The proposed method was tested on 164 fax cover pages. The experimental results show that the proposed method works well on the various types of fax cover pages.