AIRVL     Research Projects: Optical Character Recognition

[ AIRVL Home] [ People ] [Research] [ Facilities]

One of the very recent results of the group is a method called shape metamorphosis for object recognition (signatures, faces, cursive script, etc). Shape metamorphosis is a new physics-based technique that performs accurate recognition of semi-rigid or deformable objects (minimum training is required). The innovation of the approach is that it involves a segmentation method that locates corners and some key low curvature points. Due to this segmentation strategy, the proposed method can handle collectively cursive words, hand-drawn line figures, signatures, and deformable objects. The proposed system uses to its advantage the intuitive fact that two shapes (test and reference) which are similar don't have to go through an extensive metamorphosis in order for one to assume the shape of the other. Thus, the degree of morphing between a test shape and a reference shape can be used as a shape-matching criterion. The proposed recognition paradigm is invariant to translation, rotation, and scaling. The recognition results in large image databases are impressive.

Signature Verification

Automatic signature verification is a well-established and active research area with numerous applications. In contrast, automatic signature identification has been given little attention although there is a vast array of potential applications that could use the signature as an identification tool (ATM access, check verification, etc). Papanikolopoulos' work presents a novel approach to the problem of automatic signature identification. In contrast to traditional approaches that are based exclusively in an on-line or off-line system, the proposed approach uses a hybrid system. Methods based on hybrid systems not only have a definite advantage in terms of information quantity and diversity but they are also conveniently supported by technological advances in the area of hand-held communicators. The proposed method capitalizes upon the conclusions of past experimental efforts that the majority of signers have distinctive and stable signature patterns. Instead of a one-step processing model, it adopts a pyramidal processing technique by utilizing both a general and a detailed structure processing module. The general structure module has already been implemented. It deals successfully and efficiently with the bulk of the signature patterns and is intended to yield to the detailed structure module (a database of 600 signatures is used). The method deals successfully and efficiently with the bulk of the signature patterns and is intended to yield to the detailed structure module only for the problematic cases. It features a parallel physics-based algorithm, the revolving active deformable model, which affords significant discriminating power. The detailed structure module is almost complete. Its perspective role is to analyze the difficult signature images in sufficient detail by introducing the use of image metamorphosis as a powerful discriminating tool of highly variable patterns. Image metamorphosis is being implemented by following a physics-based approach. Initial experimental results show recognition rates around 85% in the database of 600 signatures.

Copyright: © 1998-2001 by the Regents of the University of Minnesota
Department of Computer Science and Engineering. All rights reserved.
The University of Minnesota is an equal opportunity educator and employer.