Automatic seal verification by evaluating positive cost

Seals instead of signature for person identification are widely used in oriental countries. Several techniques for automatic verification of seal imprints have been proposed, but all of them have dealt with this problem as a general pattern matching problem. In other words, even if it exchanges an input pattern and a reference pattern, evaluated cost does not change. In this paper, it is shown that the seal verification should not be treated as a general pattern matching problem by clarifying a specific characteristic of this problem. Two kinds of costs, called the negative cost and the positive one, may be defined in the correlation with the input and the reference patterns. A genuine imprint may produce only the negative cost to the pattern, and a forgery one may produce both costs to the pattern. Therefore, correlation value should not be calculated by definition of usual correlation. Based on the specific characteristic, a new approach for seal verification is presented. Experiments using both binary reference images and 3D reference ones provide the verification ability for the approach proposed