Abstract: A new human ear recognition approach, based on statistic features of local information, is proposed. A human ear gray-scale image is divided into several sub-regions. The assorted features of each sub-regions are abstracted. Feature of sub-regions are joined into a feature vector to build human ear feature vector, thus gives a comprehensive description about the structure and local information of human ear images. The pattern classification is implemented by applying the nearest neighbor classifier. Three different feature extraction methods are adopted and USTB human ear database are applied to testing this algorithm. The experimental results show that the recognition rate is improved more than 30% compared with global information and the method based on statistics feature of local information is proved effective.

Key words: ear recognition, local information, statistic features, Fourier transform

摘要： 提出了一种基于局部信息统计的人耳识别方法。该方法将一幅人耳图像分成若干个子区域，分别提取每个子区域的分类特征，将各个子区域的特征串联为一个特征向量构筑人耳特征矢量，更加全面描述了人耳图像的局部与结构信息，应用最近邻分类器进行模式分类。采用三种不同的特征提取方法，以USTB人耳图像库对算法进行测试，实验结果表明，与全局信息比较同种方法识别率提高30%以上，验证了局部信息方法的有效性。

关键词: 人耳识别, 局部信息, 统计特征, 傅里叶变换