Dynamic Ensemble Selection for Off-Line Signature Verification

Abstract

AbstractAlthough not in widespread use in Signature Verification (SV), the performance of SV systems may be improved by using ensemble of classifiers (EoC). Given a diversified pool of classifiers, the selection of a subset to form an EoC may be performed either statically or dynamically. In this paper, two new dynamic selection (DS) strategies are proposed, namely OP-UNION and OP-ELIMINATE, both based on the K-nearest-oracles. To compare ensemble selection strategies, a hybrid generative-discriminative system for off-line SV system is considered. Experiments performed by using real-world SV data, comprised of genuine samples, and random, simple and skilled forgeries, indicate that the proposed DS strategies achieve a significantly higher level of performance in off-line SV than other well-known DS and static selection (SS) strategies. Improvements are most notable in problems where a significant level of uncertainty emerges due a considerable amount of intra-class variability.KeywordsTest VectorDynamic SelectionAverage Error RateMajority Vote RuleEnsemble SelectionThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.