Parallel Design For Sweat Pore Recognition Algorithm Based On Embedded GPU

Fingerprint matching and recognition is one of the most popular biometrics.With the development of photoelectric technology,performance boost for fingerprint capture devices makes the extraction of high resolution fingerprint features possible,such as sweat pore.The pore recognition is one basic step for fingerprint matching,but how to achieve the effective algorithm with application into embedded system is a big problem because of its comparatively complicated calculation.However,recently,the embedded multi-core heterogeneous processor integrated with GPU provides more computing power for the application of the sweat pore recognition algorithm in mobile platform.Thus,this thesis studies the parallel software design of accelerated sweat pore recognition algorithm.In this thesis,after the study of general parallel computing of GPU,combined with the specific sweat pore recognition algorithm,analysis of parallelism.Then this thesis makes the application of algorithm in embedded heterogeneous system,and improves the efficiency of the recognition of sweat pore by using multiple optimization methods,promoting the application of fingerprint matching algorithm in embedded field.The contents and the achievements of this thesis are listed as follows:(1)This thesis studies on the recognition algorithm of sweat pore deeply and implement GPU parallel analysis on every part,such as sweat pore extraction,valley line structure characteristics extraction,sweat pore features match,and makes it sure that some modules are suitable for GPU parallel optimization,such as image contrast enhancement,gaussian differential filter,fingerprint direction field calculation,sweat pore neighborhood valley line structure extraction,sweat pore matching,and proposes corresponding parallel scheme.(2)This thesis makes hybrid programming of OpenCV and CUDA,by using Qt Creator integrated development environment and qmake build tools.This thsis puts forward the parallel implementation of sweat pore matching algorithm in GPU and improves the utilization of processor resources and the computational efficiency of the program by using various optimal means,such as dynamic parallel,zero copy memory.It shows that parallel algorithm can achieve up to 180 times acceleration in a heterogeneous embedded system.(3)This thesis designs GUI applications based on Qt,and at the same time makes the extraction of MySQL database for image data and SQL language connection in Qt environment come true,in order to manage and store image feature data effectively and makes it convenient to experimental analysis.