快速三维人耳提取与识别

Fast 3D Ear Extraction and Recognition

针对现有三维人耳提取与识别算法中存在处理时间长、识别率低的问题,提出一种快速三维人耳提取方法和2种三维人耳识别方法.三维人耳提取时,使用不变特征迭代最近点算法使人耳与平均耳对齐,完成位置和姿态的归一化,然后用掩膜提取出三维人耳.第一种三维人耳识别方法结合人耳深度和曲率信息,采用主元分析算法进行降维,然后用最近邻分类完成识别;第二种三维人耳识别方法则使用不变特征迭代最近点算法对齐测试耳与原型耳,利用配准误差完成人耳识别.实验结果表明,第一种人耳识别方法识别率较高、计算速度很快,第二种人耳识别方法可达到很高的识别率.

The main drawbacks of existing 3D ear extraction and recognition algorithms are their long processing time and low recognition rate. In this paper, a novel approach for fast 3D ear extraction and two approaches for 3D ear identification are proposed. For the ear extraction, the ear pose and position are normalized by aligning ear to the mean ear by iterative closest point using invariant features （ICPIF） algorithm. A mask is finally used to extract the 3D ear. In the first 3D ear identification approach, ear is represented by a combination of range image and curvature image. Principle component analysis is then adopted to reduce the dimensionality, followed by the nearest neighbor （NN） algorithm for ear recognition. In the second 3D ear identification approach, the ICPIF algorithm is used to align the probe ear and gallery ear. The registration error is used for ear recognition. Experimental results show that our first ear identification approach has a relatively good recognition rate but a very fast computing speed, and our second approach could achieve a very high recognition rate, but less computationally efficient compared with the first one.