基于混合测地线区域曲线演化的虹膜定位方法

Iris segmentation based on hybrid geodesic region-based curve evolution

由于可见光移动端虹膜识别中图像受光照不均、镜面反射等因素干扰,可能损失部分虹膜的结构信息和边缘信息,造成难以准确定位虹膜区域。为解决上述问题,提出一种基于混合测地线区域曲线演化的虹膜定位方法。该方法首先在图像预处理中采用拉格朗日插值法消除光斑;然后采用基于HOG特征和级联分类器的虹膜检测模型初步定位虹膜区域;再根据灰度梯度信息和拟合圆的方法确定虹膜内边缘;最后采用混合测地线区域曲线演化法定位虹膜外边缘。为测试算法的有效性,采用MICHE-I虹膜图库中由iPhone5采集的754张虹膜图像进行测试,并且提出一种虹膜外边缘定位准确度的定量评价方式。测试结果表明,方法在包含干扰的图像中虹膜定位成功率达到93.3%,虹膜边缘定位错误率为0.08,满足移动端虹膜识别的实用需求。

Due to the various interferences of uneven illumination and specular reflection in the mobile iris recognition of visible light, the structural or marginal information of some irises may be lost, making it difficult to accurately locate the iris region in the images. In order to solve the above problems, an iris localization method based on hybrid geodesic region curve evolution is proposed in this paper. Firstly, the Lagrange interpolation method was used to eliminate the light spot during image preprocessing. Then the iris region was preliminarily located by a model based on HOG and cascade. By the fitting circle method, the inner edge of the iris was determined according to the gray gradient information. Finally, the iris outer edge was detected by the hybrid geodesic region curve evolution. To test the effectiveness of the proposed method, 754 iris images collected by iPhone5 in the MICHE-I iris gallery were used and a quantitative evaluation method for the accuracy of the outer iris edge position was also proposed. The test results show that the localization success rate of the noisy iris image is 93.3%, and the error rate of iris outer edge position is 0.08, which are more suitable for the practical needs of the mobile iris recognition.