On are included in the Trichostatin A web on-line database [38, 39]. This dataset has been collected in Spain and is one of the largest on-line corpuses. ?Different signature databases have emerged during past ICDAR and ICFHR conferences. The NISDCC dataset is a database used for the Signature Competition during the ICDAR 2009. It was collected and processed by the Netherlands Forensic Institute. This corpus contains the off-line and on-line versions of the same signatures. In this work we have used the evaluation corpus which comprises 100 users, with 12 signatures per writer and 6 forgeries per signature. In total, such a corpus has 1953 signatures for both on-line and off-line datasets. [40, 41]. ?The two sub-corpuses from the on-line SUSIG database have also been considered. The Visual sub-corpus contains 94 available users, with 20 genuine signatures, acquired in two sessions, and 10 forged signatures: 94 ?20 = 1880 and 94 ?10 = 940 genuine and forged files. In total, 2820 available signatures. The Blind sub-corpus consists of 88 users with 8 or 10 genuine repetitions and 10 forged signatures per user. These sub-corpuses contain in total 820 genuine and 880 forged available files [42]. This database was collected at the Sabanci University in Turkey. ?The donors of the publicly available on-line SVC2004 signature database are mainly Chinese people, who are used to writing in English. This database is different to the others because there are Chinese and English style signatures included. This database is divided into two subsets: Task 1 and Task 2. Each buy (-)-Blebbistatin subset contains 40 users with 20 genuine and 20 forged signatures per user. Considering both subsets, all of the databases contain 2 ?40 ?20 = 1600 and 2 ?40 ?20 = 1600 genuine signatures, captured by a multi-session protocol, and forged signatures. As this work considers only Western signatures, the Chinese signatures were omitted, leaving 40 Western users [43]. These databases were collected in countries located in western, central and eastern Europe. To study the dependence of the lexical and morphological features as a function of the donors’ geographical area, a geographical region has been assigned to each dataset. In this way, the more occidental databases, i.e. the signatures included in the GPDS and MCYT databases have been grouped. As such, we have labeled both of these databases with the name DB1. The nomenclature and label assigned to each dataset are shown in Table 1. As we do not know whether there is a truly Western style in the SVC dataset, such a style will be named non-native style” in this work. All experiments were performed with all databases.Table 1. Database classification according their geographical area. Database GDPS and MCYT NISDCC SUSIG SVC doi:10.1371/journal.pone.0123254.t001 Label DB1 DB2 DB3 DB4 Style West Europe Central Europe Eastern Europe Non Native StylePLOS ONE | DOI:10.1371/journal.pone.0123254 April 10,5 /Modeling the Lexical Morphology of Western Handwritten SignaturesTherefore, summing up the five datasets, the lexical morphological features have been extracted from 881 + 330 + 100 + 94 + 88 + 40 = 1533 different signers.MethodThe method we use to characterize the lexical and morphological parameters depends on the feature properties. In this study, they are divided into three kinds: shape features (e. g. the signature envelope); discrete features (e.g. the number of words per line); and continuous features (e.g. the signature skew or slope). In the case of th.On are included in the on-line database [38, 39]. This dataset has been collected in Spain and is one of the largest on-line corpuses. ?Different signature databases have emerged during past ICDAR and ICFHR conferences. The NISDCC dataset is a database used for the Signature Competition during the ICDAR 2009. It was collected and processed by the Netherlands Forensic Institute. This corpus contains the off-line and on-line versions of the same signatures. In this work we have used the evaluation corpus which comprises 100 users, with 12 signatures per writer and 6 forgeries per signature. In total, such a corpus has 1953 signatures for both on-line and off-line datasets. [40, 41]. ?The two sub-corpuses from the on-line SUSIG database have also been considered. The Visual sub-corpus contains 94 available users, with 20 genuine signatures, acquired in two sessions, and 10 forged signatures: 94 ?20 = 1880 and 94 ?10 = 940 genuine and forged files. In total, 2820 available signatures. The Blind sub-corpus consists of 88 users with 8 or 10 genuine repetitions and 10 forged signatures per user. These sub-corpuses contain in total 820 genuine and 880 forged available files [42]. This database was collected at the Sabanci University in Turkey. ?The donors of the publicly available on-line SVC2004 signature database are mainly Chinese people, who are used to writing in English. This database is different to the others because there are Chinese and English style signatures included. This database is divided into two subsets: Task 1 and Task 2. Each subset contains 40 users with 20 genuine and 20 forged signatures per user. Considering both subsets, all of the databases contain 2 ?40 ?20 = 1600 and 2 ?40 ?20 = 1600 genuine signatures, captured by a multi-session protocol, and forged signatures. As this work considers only Western signatures, the Chinese signatures were omitted, leaving 40 Western users [43]. These databases were collected in countries located in western, central and eastern Europe. To study the dependence of the lexical and morphological features as a function of the donors’ geographical area, a geographical region has been assigned to each dataset. In this way, the more occidental databases, i.e. the signatures included in the GPDS and MCYT databases have been grouped. As such, we have labeled both of these databases with the name DB1. The nomenclature and label assigned to each dataset are shown in Table 1. As we do not know whether there is a truly Western style in the SVC dataset, such a style will be named non-native style” in this work. All experiments were performed with all databases.Table 1. Database classification according their geographical area. Database GDPS and MCYT NISDCC SUSIG SVC doi:10.1371/journal.pone.0123254.t001 Label DB1 DB2 DB3 DB4 Style West Europe Central Europe Eastern Europe Non Native StylePLOS ONE | DOI:10.1371/journal.pone.0123254 April 10,5 /Modeling the Lexical Morphology of Western Handwritten SignaturesTherefore, summing up the five datasets, the lexical morphological features have been extracted from 881 + 330 + 100 + 94 + 88 + 40 = 1533 different signers.MethodThe method we use to characterize the lexical and morphological parameters depends on the feature properties. In this study, they are divided into three kinds: shape features (e. g. the signature envelope); discrete features (e.g. the number of words per line); and continuous features (e.g. the signature skew or slope). In the case of th.