Online signature verification algorithms and development of signature international standards
Óscar Miguel Hurtado · Dialnet (Universidad de la Rioja) · 2011
The science of biometrics is based on discovering the identities of human beings by investigating their physical and behavioural traits. Of the many different biometric traits, i.e. fingerprint, iris, vascular, etc... the handwritten signature is still one of the most accepted techniques. Advancing progress in identification applications has led to widespread demand for new generation ID documents, such as electronic passports and citizen cards, which contain additional biometric information required for more accurate user recognition. This can be achieved by embedding dynamic signature features within the documentation. However, this would result in two significant drawbacks that must be addressed, these are: Memory Capacity and Computational Load. These problems and the increasing demand for standardized biometric verifications systems have motivated the research work performed in this Thesis. In order to achieve this, an attempt to reduce the information involved in verification processes is performed using feature selection criteria of the signature biometric data. Such reduced information content not only satisfies the memory capacity restrictions but also provides much more efficient use of the verification algorithms. In particular, two novel methods in the signature context, based on Principal Component Analysis and Hellinger Distance, are proposed here. The performance of the optimized features set obtained has been analyzed using two different verification algorithms. By reducing the sample size it has been observed that the error rates are maintained sufficiently low and the results obtained are in agreement with the current state of the art for signature techniques. It will be shown that in some cases that feature selection does not provide an adequate reduction solution, where a different strategy has been analyzed to achieve the aforementioned problems. A direct consequence of the widespread nature of biometric verification has led to demands for standardized protocols to improve interoperability. The work presented throughout this Thesis has considered current ISO/IEC signature standard data formats. It has been observed that the current compact data formats, 19794-7 Compact Format and 19794-11, do not meet the requirements of modern data formats. In particular, 19794-7 Compact Format, although having good compression ratios, has been found to imply an inadmissible loss in information. This problem has been solved by defining a new near-lossless compression data format based on lossless compression algorithms, and proposing different enhanced strategies to store signature data. This new data format achieves the same compression ratio, but without losing any relevant information. In addition, the problems found in the 19794-11CD2 regarding the lack of compression and information loss have been addressed. A new data format structure has been proposed, where the lack of compression is solved by reducing the data stored, avoiding duplicated data and providing a new singular point definition. This new structure has provided improved compression ratios, and, at the same time, carries more information. The two new data format definitions were presented to the ISO/IEC SC37 WG3 experts and accepted as the new third subformat “Compression Format” within the 19794-7 and the new committee draft for the 197974-11 CD3. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------