| Authentication is a mostly encountered basic matter in our daily life. With the development of the computer and network technology and the expansion of the physical and virtual space, so much more demands are put forward such as precision, security and practicability that the classical authentication methods are more and more limited in the various application areas. Some new authentication methods with more security and convenience need to be sought for. Under these circumstances, the biometric authentications were coming up and become quite popular.Dynamic signature verification is a biometric authentication technology using person's behavior characteristics in handwriting. To obtain the abundant patten information, the dynamic signature characteristics are widely adopted from the on-line collecting device. The special input device can provide manifold signature signals such as trajectory, pressure, obliquity et al, and these information compose the time sequence. Via the corresponding matching algorithm, the exterior features as trajectory and the interior features such as pressure, velocity, acceleration can be verified to supply references for authentication. Having the advantage of hard to forge, dynamic signature verification preserves the habit of ink-like signature behavior and has more acceptability in the application.Based on the stability analysis about the features of dynamic signature, a novel parallel computing technology with GPU device is proposed in the study of signature verification. And the features both in frequency region and time region are integrated into the algorithm, so that the theories of Multi-resolution Analysis, Dynamic Programming, Fourier Transform and Clustering can be well applied to the research on the dynamic signature verification.The main contents of the study and the contribution are listed in the following.1. With the rapid development of computer technology and VLSI technology, the graphics processor (GPU) on today's commodity video cards has evolved into an extremely powerful and flexible processor. The latest graphics architectures provide tremendous memory bandwidth and computational horsepower, with fully programmable vertex and pixel processing units that support vector operations up to full IEEE floating point precision. Architecturally, since the G80 series GPUs have been highly parallel streaming processors optimized for vector operations, with both MIMD (vertex) and SIMD (pixel) pipelines. Researchers have found that exploiting the GPU can accelerate some problems by over an order of magnitude over the CPU. These processors are capable of general-purpose computation beyond the graphics applications for which they were designed.CUDA is a parallel computing architecture and programming model developed by NVIDIA. The CUDA architecture includes an assembly language (PTX) and compilation technology that is the basis on which multiple parallel language and API interfaces are built on NVIDIA GPUs, including C (C++) for CUDA, OpenCL, Fortran, Matlab and DirectX Compute Shaders. C for CUDA uses the standard C language with extensions, and exposes hardware features to researchers for fully use of GPU resources. With the high level languages have emerged for graphics hardware, personal computer's computational power are more accessible and make high performance computing on the desktop come to true.2. A set of mathematical model is established in here according to the segments of dynamic signature sequences. A method is presented to create a polar coordinate system based on the mess centers of whole signature and the golden section under which we can extract the extrema from the feature of polar angle as signature segmenting-point sequence. Upon these special points, a signture can be splitted to segments with the relatively stable local features.3. After the segments achieved, the GPDM (GPU based Parallel Dynamic Matching) algorithm of parallel dynamic matching is presented according to the CREW PRAM parallel model for matching the time sequences among the small capacity library with dynamic programming technology. Dynamic programming is a global optimization method to match the time sequences with variable length by the evaluating function. It has various implementing algorithm applied to the pattern recognition area and it is also a primary method in dynamic signature verification. Since there are specialities of the dynamic signature time sequence with well stability in global scale and variability between phases, the segment's correlativity is much more important to the verification course rather than some special points's matching result. To reduce the computing quantity, the GPGPU is adopted to accelerate the performance in dynamic matching algorithm. Evaluated using the JLU-DHSDB 2.0 signature database, The GPDM algorithm resulted in 4.10% EER.4. A method for features extraction and dimension reducing was presented here based on MRA, to promote the performance of the dynamic programming algorithm furthermore. The wavelet transform is a synthesis of ideas that emerged over many years from different fields, such as mathematics and signal processing. Generally speaking, the wavelet transform is a tool that divides up data into different frequency components and then studies each component with a resolution matched to its scale.Using wavelet decomposition, the approximate coefficients of signature at high scale can be applied to obtain the extrema based on the polar angle feature. As the result, the much more eligible stable segments can be obtained accordingly and then we can calculate the similarity at lower scale coefficients. By wavelet decomposition the dimension of signature sequence can be reduced more efficiently to save the quantity of computing operations ulteriorly in the following verification phase.5. An algorithm of Iteration based Hidden Hierarchical Clustering (IHHC) was proposed. Clustering was performed based on the features extracted from frequency region with Fourier Transform, which can be used to optimize the signature template database. The signatures written by enrollee in a short period contain more redundancies which are not suit for the constructing the reference template effectively and that can induce the false rejection in next verification after a long period. Using the Clustering method can optimize the signature template on the one hand and on the other hand it can scout the trend of the man's handwriting and trace the new feature which never occured before. Thus it can promote the performance more efficiently and increase the verification precision.6. Evaluation databases used in this dissertation is JLU-DHSDB2.0 which has been established specially by the Dynamic Signature Verification Team of Jilin University. There are 127 categories in the database corresponding to 127 individual donor's original dynamic handwritten signature data. In each category it contains at least 20 skilled genuine signatures, additionally some of the enrollee signed another 10 genuine signatures after 3 weeks. There are also 60 skilled forgeries and 10 counterdrawing forgeries in the every category. This database is used to evaluate the stabilities of signature and accuracy of the authentication algorithm widely.
|
PDF Full Text Request