N-Latticed Bounding Box-Based Signature Pattern Recognition by Accumulated ASCII Difference

摘要

Recently, the efficiency of user recognition method is getting more importance as growing the needs for personalized service in various human-machine interaction systems. One of the representative methods for user recognition is using the signature of user, a kind of behavior-basedbiometric method. But most of previous online signature recognition requires a burden of computing cost. As a result, it is not well suited for various simple applications. To overcome this drawback of previous online signature recognition, simple grid-based signature representation methodwas developed, named as input window method. But, the input window method has basically two limitations; (1) no consideration for signature normalization and (2) noise sensitivity from the hard decision-maker such as exact string comparator. In this paper, a novel N-latticed boundingbox-based signature pattern recognition method is developed by using the accumulated ASCII difference as comparator. The experiments with 8 people and 17 lattices show that signature could be normalized properly in the view of size and noise could be handled by showing 91.7% averaged successratio. In addition, it still has the advantage of efficiency by the simplicity of data representation since the generated signature pattern is simple 1D data by the help of N-lattice lines.