指纹图像的旋转校正与分类

Rotation Correction and Classification of Fingerprint Image

摘要: 为了提高在大容量指纹数据库中指纹识别率的速度和正确率，也为了提取出更多的细节特征，提出了一种旋正图像并使用中心点特征进行指纹分类的方法。首先，根据指纹图像的最小外接椭圆和矩形，获得旋转角度并对指纹进行仿射变换以校正图像；然后，针对现有Poincare index方法存在伪点的问题，通过对其改进实现中心点的确定，并总结人眼识别的经验对采集到的中心点去噪；最后，利用中心点的绝对方向角度、螺径等细节特征对指纹进行分类。对于无法分类或分类不正确的图像，采取二次匹配保证正确率。在从电容式指纹传感器FPC1011F采集到的400幅图像上，分类的准确率为91.25%，实验结果验证了该方法的有效性和鲁棒性。

Abstract: In order not only to improve the speed and accuracy of fingerprint recognition in the large-capacity database, but also to extract more detailed features, this paper presents a classification algorithm by using the features of cores, which are extracted from a corrected fingerprint. In the first phase, according to the smallest external ellipse and rectangle, the declining fingerprint is corrected by the affine transformation. In the second phase, to heighten anti-noise capability of traditional Poincare index, an improved algorithm is proposed, and then false points are denoised by summarizing the human recognition experience. Finally, the absolute direction, the diameter of screw and other features are the basis for fingerprint classification. 400 fingerprint images collected by FPC1011F fingerprint sensor are used for an experimental test, and the accuracy rate on classification is 91.25%. The experimental results show its effectiveness and robustness.