Improving performance of HMM-based off-line signature verification systems through a multi-hypothesis approach

Batista, Luana; Granger, Éric; Sabourin, Robert

doi:10.1007/s10032-009-0101-0

Cited by 17 publications

(19 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…With OP-UNION and OP-ELIMINATE, the search for the K-nearest neighbors is done by using the 362 output labels provided by all 100 SVMs; while with KNORA-UNION and KNORA-ELIMINATE, only the 363 SVM input subspace providing the lowest error rates on DB The dynamic selection strategies proposed in this paper are compared as well with two reference systems 385 proposed in our previous work, that is, (i) a traditional generative system based on HMMs [4] (refered in this 386 paper as baseline system), and (ii) a hybrid system based on the static selection of generative-discriminative 387 ensembles [35]. Both systems are briefly described in Section 5.…”

Section: Classifier Ensemble Selection 358mentioning

confidence: 99%

“…Two different off-line signature databases are used for proof-of-concept computer simulations: the Brazil-261 ian SV database, used by our research group [4,6,23,35], and the GPDS database [12], used by other 262 researchers [5,36,37,38]. While the Brazilian SV database is composed of random, simple and skilled 263 forgeries, the GPDS database is composed of random and skilled forgeries.…”

Section: Off-line Sv Databases 260mentioning

confidence: 99%

“…4 For complete and recent surveys of off-line SV, the reader is referred to [1,2]. The Hidden Markov Models 5 (HMMs) [3] have been successfully employed for classification due the sequencial nature and variable size 6 of the signature data [4,5,6,7]. In particular, the left-to-right topology of HMMs is well adapted to the 7 dynamic characteristics of European and American handwriting, in which the hand movements are always 8 from left to right.…”

Section: Introductionmentioning

confidence: 99%

“…Given a digi-3 tized signature, an off-line SV system typically performs preprocessing, feature extraction and classification. 4 For complete and recent surveys of off-line SV, the reader is referred to [1,2]. The Hidden Markov Models 5 (HMMs) [3] have been successfully employed for classification due the sequencial nature and variable size 6 of the signature data [4,5,6,7].…”

Section: Introductionmentioning

confidence: 99%

“…To validate the proposed SV system, proof-of-concept experiments are carried out on real-world signature 39 data from two datasets, namely, the Brazilian SV database [4] (comprised of genuine samples, and random, 40 simple and skilled forgeries) and the GPDS database [12] (comprised of genuine samples, and random and 41 skilled forgeries). The performance of the generative-discriminative ensembles formed with the proposed 42 dynamic selection strategy is compared to that of other well-know dynamic and static selection strategies, 43 with a traditional system based on HMMs, and with other relevant SV systems found in the literature.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

Dynamic selection of generative–discriminative ensembles for off-line signature verification

Batista

Granger

Sabourin

2012

Pattern Recognition

Self Cite

View full text Add to dashboard Cite

In practice, each writer provides only a limited number of signature samples to design a signature verification (SV) system. Hybrid generative-discriminative ensembles of classifiers (EoCs) are proposed in this paper to design an off-line SV system from few samples, where the classifier selection process is performed dynamically. To design the generative stage, multiple discrete left-to-right Hidden Markov Models (HMMs) are trained using a different number of states and codebook sizes, allowing the system to learn signatures at different levels of perception. To design the discriminative stage, HMM likelihoods are measured for each training signature, and assembled into feature vectors that are used to train a diversified pool of two-class classifiers through a specialized Random Subspace Method. During verification, a new dynamic selection strategy based on the K-nearest-oracles (KNORA) algorithm and on Output Profiles selects the most accurate EoCs to classify a given input signature. This SV system is suitable for incremental learning of new signature samples. Experiments performed with real-world signature data (comprised of genuine samples, and random, simple and skilled forgeries) indicate that the proposed dynamic selection strategy can significantly reduce the overall error rates, with respect to other EoCs formed using well-known dynamic and static selection strategies. Moreover, the performance of the SV system proposed in this paper is significantly greater than or comparable to that of related systems found in the literature.

show abstract

Section: Classifier Ensemble Selection 358mentioning

confidence: 99%

Section: Off-line Sv Databases 260mentioning

confidence: 99%