指节纹图像感兴趣区域提取与特征识别算法

ROI extraction and feature recognition algorithm for finger knuckle print image

为降低特征提取的工作量同时提高识别的准确性,设计了一种指节纹图像感兴趣区域(ROI)提取方法。首先,对采集到的图像进行预处理操作,分离并旋转定位四指图像。其次,通过手指图像梯度分布关系,计算负梯度的极值,并统计图像中每一行的负梯度极值个数来确定近指节纹的ROI区域。最后,通过局部二值模式(LBP)直方图相似性来验证所提取的ROI区域的准确性。实验证明,在采用本文方法所建立的指节纹ROI数据库中,分类准确率达到了100%。

When the images of finger knuckle print are used as identity characteristics,there exist obvious differences in the hand position and illumination condition due to the non-contact hand shape acquisition instrument.In order to reduce the workload of the feature extraction and improve the recognition accuracy,a new method for Region of Interest(ROI)extraction of finger knuckle print image is proposed.First,the preprocessing of the image is executed,and the finger images are separated and rotated.Then,the extrema of the negative gradient are calculated according to the distribution relationship of the image gradient,and the ROI of the near knuckle print is confirmed by calculating the number of extrema of negative gradient in each line of the image.Finally,Local Binary Pattern(LBP)operators are used to verify the accuracy of the ROI.Experiment results show the ROI database built by the proposed method is accurate and recognition rate can reach 100%.