An On-line Handwritten Signature Authentication Technology For Mobile Devices

With the wide spread of mobile devices, safe and efficient end-user authentication for mobile devices is becoming particularly urgent. Based on the analysis of needs of signature verification on mobile devices, and the current status of online signature verification technology at home and abroad, we propose to use a small sample, one-class classification method, low complexity algorithm authentication method for mobile devices. The main work is as follows:We relies on the Android phone built-in software for signature data collection, and abandons the sole use of external F_Table handwriting platform and dedicated stylus in the traditional methods. The software is designed to have function of saving and clearing, can collect and store in real time the sample of the user's signature to the SD card of the Android phone, in order to facilitate the users' capturing and use on mobile devices.As the required computation for the identity authentication of mobile equipments should have characteristics of low complexity, we use the feature extraction method based on DCT frequency domain feature analysis and singular value feature fusion, and DCT frequency domain feature analysis to compress the feature vector in the low frequency domain to save a lot of computing time, and implement the feature dimension reduction for the characteristic matrix effectively by the singular value decomposition.For the characteristics of authentication on mobile devices with limited number of target samples, by using the limited samples applicable for the SVDD classifier, and the advantages of modeling by one-class classification, this paper uses SVDD to set up classification authentication model, and uses the grid search, particle swarm optimization respectively to optimize the kernel parameters. By using the experimental results of contour map of grid parametric search, and the particle swarm fit curve, the paper shows the optimal parameters searched, and substitutes it into the SVDD classification authentication model, to obtain a high accuracy rate, indicating grid search method and particle swarm optimization method have advantages of high speed and good learning precision.Finally, this paper makes on-line handwritten signature authentication experiments on the PC and analysis to the related data, which provides a basis for the effective algorithm for its running on mobile devices in the future.