摘要
为了提高多姿态人耳的识别准确率,提出了一种结合梯度方向直方图(HOG)特征和监督保局投影的人耳识别方法.人耳图像被划分成重叠的子区域,在每个子区域中计算局部HOG特征.所有的局部HOG特征构造成包含丰富信息的高维向量,高维向量映射到流形空间利用监督保局投影进行鉴别分析,获得强鉴别力的特征.利用最小欧氏分类器进行分类识别.结果表明,本文方法提高了大姿态人耳的识别率.稀疏的子区域表示和局部HOG特征能在一定程度上克服姿态变化造成的对齐误差,对角度变化有很好的鲁棒性.
To improve recognition accuracy of multi-pose ear, a novel method is presented which combines histograms of oriented gradients （HOG） feature with supervised locality preserving projection （SLPP）. The ear image is first represented by a set of overlapped sub-regions,and then the local HOG features are computed on each sub-region on account of its invariance to local geometric transformation. The local HOG features are constructed as a high dimensional vector containing ample information, and then the high dimensional vector is mapped to manifold space to acquire stronger discriminant feature using SLPP. Finally, the minimum European distance classifier is used for recognition. Experimental results show that a higher recognition rate is achieved by the proposed method than by other related methods. The feature representation of the sparse sub-region and SLPP can compensate for the alignment errors caused by pose variations, and the recognition method can efficiently improve the robustness of pose variation.
出处
《西安工业大学学报》
CAS
2013年第11期878-883,共6页
Journal of Xi’an Technological University
基金
陕西省教育厅专项科研计划项目(2010JK595)
关键词
梯度方向直方图
人耳识别
监督保局投影
局部特征
histograms of oriented gradients （HOG）
human ear recognition
supervised locality preserving projection（SLPP）
local features
