Palmprint recognition research and DSP Implementation
|School||Xi'an University of Science and Technology|
|Course||Signal and Information Processing|
|Keywords||Palmprint recognition Two-stage match Improved Sobel operator False Acceptance Rate False Rejection Rate DM642 Optimization|
Palmprint recognition biometric technology is an emerging area of ??technology, it is an effective means manpower Pocket palmprint image information to identify which are from the same only the palms, which palmprint image is from a different hand, and as a basis to identify the user's identity. Currently palmprint identification system in a variety of applications of biometric systems account for a small portion of the amount in the domestic civil palmprint recognition system application market is almost empty. Well developed compact, inexpensive, short development cycle, easy to maintain, to meet medium / low-level security requirements palmprint identification system to promote the use in civilian areas would be a meaningful research. Mature core algorithm palmprint identification number, identification methods based on statistical features are FFT transform, wavelet transform, K_L transformation. This view of the FFT transform principle is simple and easy to achieve, thus using palmprint recognition based on the FFT method, the drawback is the recognition rate compared palmprint recognition based on wavelet transform is much less. This article is based on the core idea palmprint identification radius of eigenvalues ??palm rough match to image 8 FFT transform spectrum of energy and direction for the characteristic values ??for fine matching two-stage matching thought. In the preprocessing stage, according to the principle of morphological palmprint image positioning ideas, algorithms and achieved its advantages and disadvantages of this algorithm are described; based on \one-dimensional linear interpolation processing, advantages and disadvantages of doing the simplification algorithm is described in detail, and through the decomposition of the rotation angle of the method effectively avoid carrying large angle when rotating the image pixel loss problem. This article uses a template based on eight directions Sobel operator to make an improved image edge extraction, get a good edge effect, the algorithm extracts the edge-to-edge noise suppression when the effect is prominent, edge sharpening characteristic distinctive features, proposed the use of improved Sobel operator Sub-do image enhancement methods. In the matching stage, were used for the classical Sobel algorithm combines bimodal threshold extract palmprint image ridge two values ??do FFT transform, the improved algorithm combined with single threshold Sobel extract palmprint image ridge two values ??do FFT, Sobel image enhancement for improving the processed image directly to do FFT, these three different pretreatment methods, through the preprocessing algorithm matches the three test methods a system to identify the lowest rate, the lowest rate of false acceptance, false rejection the highest rate; method two highest recognition rate system, the highest rate of false acceptance, false rejection rate is the lowest; method three of the recognition rate between the two. Practical applications, according to the characteristics of three methods of identifying the three bonding systems application environment select the most appropriate method. DM642 DSP hardware platform is a development board, writing in the code, the control loop pipelining, register configuration and so do the optimization program run time from 7.9s down to 1.8s, optimization effect is obvious, to meet real-time requirements.