摘要
针对人耳角度变化引起识别率下降的问题,提出一种结合Log-Gabor滤波和正交局部保持投影(OLPP)的人耳识别方法。首先采用Log-Gabor对图像进行滤波来提取多尺度多方向的人耳纹理特征;然后在局部保持投影的原始优化问题中增加正交约束条件,迭代计算出一组具有正交性最优映射向量,约简了丰富的Log-Gabor特征,并保留了人耳非线性流形子空间与距离有关的结构信息和重构样本;最后用最小欧氏距离分类器进行分类识别。对比相关的方法,该方法提高了姿态人耳的识别率。实验结果表明该方法能良好地表征姿态人耳,对角度变化具有很好的鲁棒性。
Aiming at the decline in recognition rate caused by the variation of human ear angle, we propose in this paper a novel ear recognition method which is based on Log-Gabor filter and orthogonal locality preserving projection （OLPP）. First, the multi-scale and multi-orientation texture feature of ear image is extracted from the image using Log-Gabor filter. Then the orthogonal constraint conditions are added to the primitive optimisation problem in regard to locality preserving projection, and a set of projection vectors with orthogonal optimum is calculated through iteration. Abundant Log-Gabor features are reduced, and the structure information and reconstruction sample of nonlinear submanifold space of the ear related to distance are preserved. Finally, the minimum Euclidean distance classifier is applied in classification and recognition. In contrast to the correlated method, the proposed method improves the recognition rate of pose ear variation. Experimental result shows that this method can well represent multi-pose ear image, and is robust to the variation of ear angle.
出处
《计算机应用与软件》
CSCD
北大核心
2014年第10期172-175,共4页
Computer Applications and Software
基金
陕西省教育厅专项科研计划项目(2010JK595)
