基于多尺度Canny算子的人耳几何特征提取与识别

Ear geometric feature extraction and recognition based on multiscale canny

针对人耳特点提出了一种基于耳廓边缘几何结构的识别方法,在结合多尺度Canny算子检测出单像素耳廓边缘的基础上,通过提取耳轮边缘上的若干关键距离来描述耳轮形状,提取耳甲腔边缘上的角点区域来描述内耳结构,最终利用耳轮形状特征向量及内耳结构特征向量实现分类识别。特征向量的表示分别以长轴和质心作为参照标准,保证了其具有平移、旋转和缩放不变性。在选取的人耳图像库上实验取得了较高的识别率,同时,对光照条件变化具有一定的鲁棒性,实验结果表明该方法的有效性以及利用人耳图像进行身份识别的可行性。

A novel method of ear recognition based on ear geometrical structure is proposed. First,Canny operators with different scales are applied to deteete the ear contour edge. Second, to describe the shape and structure of ear, several key dis tances are extracted from the edge of helix,and comer point areas are detected from the edge of coneha. In the end helix shape feature vector and concha structure feature vector are utilized in the process of classification. By using longest axis and eentroid as frames of reference,the feature vectors extracted in the method are invariant to parallel move,rotation and scale. The recognition rate on selected ear image database is very high, which is robust to changes in illumination. The experimental results confirm the effectiveness of this approach and prove the feasibility of using ear as biometrics for person identification and authentication.