基于仿生模式识别的多镜头人脸身份确认系统研究

提出了一种利用仿生模式识别原理以多镜头信息融合的人脸身份确认方法.讨论了以多镜头与多次采样建立仿生模式识别多权值神经元网络的基本理论.并介绍了以三个镜头五次采样作训练样本的实验系统,以及人脸预处理过程步骤.实验结果表明在确保无误认的前提下,正确确认率达96%,漏认率(即拒认率)为4%.实验中并对比显示了增加神经元网络复杂度提高识别效果的作用.

On Cyclic Correlation Matched Filtering

By employing spectral correlation analysis method,the problem of optimally filtering the cyclostationary signals is discussed in this paper.Based on the max output SNR criterion,the analytic expression for the cyclic correlation matched filter (CCMF) is derived,which is just the same as the conventional matched filter except for different signal models.But since the filtering performance for CCMF is primarily related to the selected cycle frequency,which is not unique for most cyclostationary signals,single cycle CCMF shares the disadvantage of utilizing the incomplete signal information.CCMF bank utilizing multi cycle frequencies is therefore studied and an optimum structure for the CCMF bank is developed,also based on the max output SNR criterion.Simulations on AM signals and BPSK signals are also performed to determine the performance advantage between the spectral correlation method and cyclic correlation matched filtering method.Simulation results verify the performance for the proposed methods.