摘要
本文给出了一种基于仿生模式识别理论的虹膜识别算法,该识别算法首先进行虹膜的边界定位、归一化及有效区域的选择,并采用Daubechies-4小波变换及奇异值分解的方法进行虹膜特征的提取,最后利用基于超香肠神经元的双权值神经网络对同类虹膜样本构建出了其对应的高维闭合空间覆盖面。在中科院自动化所的虹膜数据库(CASIAver.1.0)中的仿真实验表明,本文所提算法取得了良好的正确识别率,并且对于未参与训练的待识别虹膜样本能具有较高的拒识率。
an iris recognition algorithm based on the biomimetic pattern recognition (BPR) theory is presented in this paper. First, edge location, normalization and the selection of the effective location is applied to the iris image, and Daubechies-4 wavelet transform (WT) and singular value decomposition (SVD) are used to extract the feature of it, then hyper sausage neuron is adopted in double-weights neural network to construct the high-dimensional closed covering space of the same class iris samples. Tested samples is recognized or rejected by judging whether tested samples are covered with any hyper sausage chain. Simulation results in CASIA iris database 1.0 show that the proposed method can achieve high correct recognition rate, and maintain high rejection rate with the iris samples untrained.
出处
《电路与系统学报》
CSCD
北大核心
2012年第2期43-48,共6页
Journal of Circuits and Systems
基金
国家自然科学基金项目(61072127)
广东省自然科学基金项目(10152902001000002
No.07010869)
广东省高等学校高层次人才项目(粤教师函[2010]79号)
广东高校优秀青年创新人才培养计划项目(粤财教[2008]342号)
广东省产学研合作项目(2009B090300416)
五邑大学青年科研基金(Q948)
关键词
仿生模式识别
虹膜识别
超香肠神经元
小波变换
奇异值分解
biomimetic pattern recognition
iris recognition
hyper sausage neuron
wavelet transform
singular value decomposition