Abstract: In the extraction of iris features, most methods directly perform some transformation on the iris-normalized enhanced images. To reduce the iris feature dimension and ensure the recognition accuracy, this paper proposes the method of normalizing iris radial folding block and circumferential cycle block and then carrying out haar wavelet transform, which reduces the sensitivity of the iris region to noise and ensures that effective iris features are not neutralized while reducing the iris feature dimension. In order to further overcome the sensitivity of the iris recognition to rotation, this paper uses a periodic wavelet transform to extract high frequency information. Finally, using the classification method of BP（Back Propagation） neural network, the high-frequency information after wavelet transform is directly used as the input of the classifier, which further improves the accuracy of iris recognition. Experiments show that the number of feature points can be as low as 120 and the correct recognition rate can reach 99.48% by using the method proposed in this paper.

Key words: iris recognition, folding block, periodic extension, wavelet transform, zero-crossing detection, BP neural network

摘要： 在对虹膜特征提取时，绝大多数方法是直接对虹膜归一化后的增强图像进行某种变换，为降低虹膜特征维度，同时保证识别效率，提出了对归一化虹膜径向折叠分块、环向周期分块再进行haar小波变换的方法，降低了虹膜区域对噪声的敏感性，在减少虹膜特征维度的同时，保证了虹膜有效特征不被中和。为进一步克服虹膜识别中对旋转的敏感性，采用了周期延拓的小波变换方式提取高频信息。最后利用BP（Back Propagation）神经网络的分类方法，将小波变换后的高频信息直接作为分类器的输入，进一步提高了虹膜识别正确率。实验表明，提出的方法特征点数低至120，正确识别率可达到99.48%。

关键词: 虹膜识别, 折叠分块, 周期延拓, 小波变换, 过零检测, BP神经网络