The Research Of Accelerated Algorithm For Fingerprint Identification Based On CUDA

The fingerprint identification algorithm based on feature point matching is the most widely used one in security field, which actualizes real-time processing by adop-ting hardware such as FPGA and DSP. However, when the fingerprint is seriously sta-ined, the hardwared system will hardly recognize the target or satisfy the requirement for real time as a result of the excessive complex algorithm. The algorithm based on optical joint transform correlation,which adopts optical system to carry out cross corr-elation for the fingerprint to be identified and the one in the library,it will be of robus-tness for the stained fingerprint. In addition, utilizing high parallel characteristic is po-ssible to actualize fast-recognition for the system, which satisfies real-time characteri-stic. But the system requires high quality environment, which makes it difficulty for t-he popularization of the system.This paper proposes a joint transform correlation algorithm based on orient filter, which carries out the algorithm through the use of image features in different orientat-ion of the fingerprint, and it will improve the accuracy for fingerprint identification. On this basis, this paper compiles the parallel processing algorithm of joint transform correlation based on GPU platform, which enhances flexibility in different occasions, and satisfies the real-time requirements at the same time. Around the acceleration for the fingerprint recognition algorithm on GPU platform, the research work for this pa-per mainly includes the following aspects:(1) Discussing the excecution mechanism, structure model of thread, the hierarc-hical model of memory and software system. Establishing the experimental platform of CUDA and designing the apportion model of thread aiming at this topic according to the testing parameter of the experimental environment.(2) Intensively stdudying the algorithm of joint transform correlation base on ori-ent filter, actualizing parallel calculation based on GPU under VC platform by adopti-ing the video card of NVIDIA GeForce GT525M. Confirming that it is feasible to ac-celerate the algorithm of joint transform correlation. (3) Intensively stdudying the algorithm of fingerprint indentification, including i-mage denoising, image segmentation and image enhancement. Discussing schemes b-ased on CPU and GPU respectively, and carrying out the acceleration performance te-st for the algorithm based on CUDA.