Fingerprint Images Parallel Algorithms Based On MMX Technology

Because of uniqueness and invariability, the fingerprint identification is becoming one of the most popular personal authentication technologies, and it has been used in many regions. With the development of the society, people demand for higher performance of the identification system, so an exact and fast fingerprint identification system is still a difficult research subject.MMX technology is developed for multimedia and communication applications, it is an enhancement of Intel Architecture. The technology adds fifty-seven new instructions, eight registers of sixty-four bits and four kinds of data types, and it affords a parallel operation method of SIMD that a single instruction deals two, four or eight data, the parallel operation can improve the computation efficiency. When using MMX code in fingerprint images processing, it can obviously improve the computation speed with the same results.After studying on preprocessing and feature extraction of fingerprint images, the paper proposes some new parallel processing algorithms based on MMX technology:(1) When calculating direction image, a parallel processing algorithms based on MMX technology is proposed, the algorithm firstly transforms the formula and then uses MMX technology to calculate in a four-parallel way.(2) When binarization, the paper quantitates direction in order to reduce many repeating calculations.(3) In the process of thinning, a parallel thinning algorithm based on MMX technology is proposed, it uses MMX to synchronously compare the neighbours of 4 X 4 with four templates.(4) In the aspect of feature extraction, the paper estimates feature by calculating across number and 8-neighbour ridge point number with MMX technology. In the post-processing, the paper identifies and eliminates some typical false feature structures.Simulation experiments have been made for all old and new algorithms. Results show that all new algorithms given in this paper can significantly improve the computation efficiency with the same results.