Abstract: Prediction of performance in Off-line Automatic Signature Verification (ASV) per signer is one of the important topics regarding to automatic verification. It could be hypothesized that the performance of a signer is related to its global stability. This way, the more stable the signer signatures, the smaller the area of its feature space is, being more difficult to get inside for an impostor. In this paper we assess the feasibility to predict the performance of a signer through his/her global stability. As in a real scenario, only the enrolled signatures are used to calculate the stability of the signer. Similarly, only these signatures are used to train two completely different off-line ASVs. Then, the performance and the stability per signer are compared. Our results suggest that there is a certain relationship between the global stability of the enrolled signatures and the performance in terms of Equal Error Rate.
Keywords: stability analysis, optical imaging, optical distortion, entropy, image segmentation, computer vision, optical flow
