Assessment of the quality of handwritten signatures based on multiple correlations

Assuring the quality of individual biometric samples is important for maintaining the discriminatory power of biometric recognition systems as biometric data of low-quality are likely to be mismatched. This paper presents an investigation into the assessment of the quality of handwritten signatures, predicting the performance or 'utility' of individual signature samples in automated biometric recognition. The prediction of utility is based on multiple correlations with static and dynamic signature features. First, the utility of handwritten signature samples from publicly available databases is assessed by comparing them with each other using commercial automatic signature verification engines. The samples are classified into four quality bins (excellent, adequate, marginal, and unacceptable quality) with totally ordered bin boundaries. Then, the correlation of multiple static and dynamic signature features with utility is analysed to find features that can be used for predicting the utility of samples. Our results show that it is possible to predict the utility of handwritten signature samples using a multi-feature vector.