OFFLINE SIGNATURE IDENTIFICATION AND VERIFICATION USING NONITERATIVE SHAPE CONTEXT ALGORITHM
Marcin Adamski, Khalid Saeed · 2009
The paper presents experimental results on offline signature identification and verification. At the f irst stage of the presented system, the binary image of the signature undergoes skeletonization process using KMM algorithm to have a thinned, one pixel-wide line, to which a further reduction is applied. For each thinned signature image a fixed number of poin ts comprising the skeleton line are selected. The recognition process is based on c omparing the reference signatures with the question ed samples using distance measure computed by means of Shape Context algorithm. The experiments were carried out using a database containing signatures of 20 individuals. For the verification process random forgeries were used to asses the system error. The main advantage of the presented approach lies in utilizing only one refer ence signature for both identification and verifica tion tasks, whereas the achieved results are comparable with respect to the systems that use several training samples per subject. The problem of automatic identity verification is c rucial for security of data and restricting access to protected resources. Traditional methods such as passwords, P IN numbers and identification cards do not provide 100% safety and are cumbersome in everyday usage. In order to improve the security together with the comfort of their appl ication the biometric methods start to gain popularity and become the fir st choice solutions in many environments, where the re is a need for personal authentication. One of the most popular behavioural biometrics that is used on everyday basis is a han dwritten signature. Despite of its drawbacks such as relatively weak pe rmanence [5] and ease of producing a forgery, the h andwritten signature received a lot of attention in both research and in commercial institutions. The ongoing development o f new algorithms for feature extraction and signature recognition result ed in systems with error rates comparable with the results received for other biometrics. This work presents an offline system ‐ recognition process is based on the analysis of sta tic signature images. An example of other approaches, where dynamics of signing activity is being considered can be found in [3]. The main aim of this work is an attempt to create m ethod for recognition of handwritten signatures usi ng only one reference sample per individual. This may be an imp ortant asset in some practical applications where t here is only one genuine signature available for comparison. The experiments that were carried out showed that described system gives good results that are competitive with approaches using several refer ence samples for each subject.