Histogram-Based Matching of GMM Encoded Features for Online Signature Verification
Abhishek Sharma, Suresh Sundaram · 2018
This paper explores a scheme for verifying the authenticity of online signatures. We make contributions with regards to (i) the encoding of features used and (ii) the matching strategy. Most works in the literature that consider comparing the temporal sequence between two signatures utilize the local features derived at each sample point of the online trace. One aspect worth investigating is on representing these features in a probabilistic framework. In this work, we use the parameters from a pre-learnt Gaussian Mixture Model (GMM) to encode the features. Secondly, with regards to matching strategy, we consider comparing a histogram between a test signature with those obtained from the enrolled signatures of a user. The histogram is generated from the GMM encoded features by partitioning the online trace of the signature into a number of segments. Experiments conducted on the publicly available MCYT-100 database suggest reduced Equal Error Rates over the traditional GMM likelihood based systems.