A similarity measurement method robust to background noise

✍ Scribed by Eiji Hasegawa; Yoshihiko Nomura

• Book ID: 102661548
• Publisher: John Wiley and Sons
• Year: 1999
• Tongue: English
• Weight: 786 KB
• Volume: 30
• Category: Article
• ISSN: 0882-1666
• DOI: 10.1002/(sici)1520-684x(199902)30:2<38::aid-scj5>3.0.co;2-s

✦ Synopsis

It is extremely difficult to extract character strings correctly and to remove noise completely when background noise exists in the character region, as with newspaper headlines or magazine covers. This paper proposes a method of similarity estimation that reduces noise by the combined use of potential-function convolution and range transformation, so that character recognition is made possible even with background noise. With the proposed method, character models are first subjected to affine transformation in the potential field generated from the input image, and then matched to the input image. Background noise is suppressed and displacements are corrected; in addition, scale changes, cuts, rotations, and so forth can be processed to a certain extent. Noiselike black pixels are then removed by using the potential image and the range image. Finally, the similarity is determined in terms of the Euclidean distance for every pixel in order to deal with displacements and local deformations.