Feature Selection Based on Genetic Algorithms for On-Line Signature Verification

Two different genetic algorithm (GA) architectures are applied to a feature selection problem in on-line signature verification. The standard GA with binary coding is first used to find a suboptimal subset of features that minimizes the verification error rate of the system. The curse of dimensionality phenomenon is further investigated using a GA with integer coding. Results are given on the MCYT signature database comprising 330 users (16500 signatures). Signatures are represented by means of a set of 100 features which can be divided into four different groups according to the signature information they contain, namely: i) time, ii) speed and acceleration, iii) direction, and iv) geometry. The GA indicates that features from subsets i and iv are the most discriminative when dealing with random forgeries, while parameters from subsets ii and iv are the most appropriate to maximize the recognition rate with skilled forgeries.