一种适于虹膜识别的特征提取算法

Effective Feature Extraction Algorithm for Iris Recognition

为克服现有基于线性变换特征提取方法中基向量非动态和参数需指定的缺陷,分析了虹膜的几何特征和识别原理,提出用独立成分分析ICA(Independent Component Analysis)方法进行虹膜特征提取,最大限度地去了除虹膜特征空间的冗余,克服了传统线性变换特征基向量非动态的缺陷;用BP(Back Propagation)神经网络进行虹膜分类,实现特征的降维和有效表示,并在自主研制的JLU-IRIS虹膜图像库中进行小样本空间实验。结果通过三种不同的识别率100%,96.5%和92.5%,表明了该算法的正确性和有效性。

In order to avoid the shortcoming of non-dynamic base vectors and specific parameters in existing feature extraction based on linear transformation, after analyzing iris geometry features and recognition principles, an iris feature extraction using independent component analysis method is proposed. The algorithm eliminates the iris feature space redundancy furthest, and overcomes the flaw of non-dynamic feature base vectors in traditional linear transformation. The iris classification with BP neural network achieves lower dimension and effective feature expression. Experiments in small sample space using self-made iris database JLU-IRIS indicate the accuracy and validity of the algorithm through three groups of different recognitions rate 100%, 96. 5% , 92.5%.