Robust Matching with Form Model for Faxed General Form Understanding
Takashi Hirano, Yasuhiro S. Okada, Fumio Yoda · Transactions of the Institute of Electronics, Information and Communication Engineers · 2002
FAXOCR is a device that reads handwritten characters on a faxed form. However, it is difficult to extract fields for character recognition when the form image includes deformation or noise. In order to cope with such a problem, special forms have been used with constraints on the format structure. This paper proposes a method of field extraction in which a general form can be read without constraints on the form structure. In the method, the feature points extracted from the form image and the feature points of the form model are matched. The deformation of the form image is estimated, and the position of the field is determined. Since the feature points can be extracted from the tables, character sequences, and illustrations which are contained in ordinary forms, the method can be applied to general forms with diversified structures. The estimation of deformation is robust to disappearance of feature points or to new appearance of excessive feature points, and the method is robust to degradation of the form image due to noise and other disturbances. The proposed method was applied to 650 general form images of 50 different kinds with deformations inherent to fax, and its effectiveness was demonstrated. The method has also been applied to field data, and performance at a practical level (form rejection rate of 3.5%) has been obtained. © 2004 Wiley Periodicals, Inc. Electron Comm Jpn Pt 3, 87(12): 60–72, 2004; Published online in Wiley InterScience (www.interscience. wiley.com). DOI 10.1002/ecjc.20130