摘要
局部方向模式(LDP)利用8个Kirsch模板与3×3局部邻域卷积得到局部的边缘梯度值,然后取绝对值并排序,最后将最大的三个梯度所在的方向信息编码成一个八位二进制数;其不足之处在于将边缘梯度求绝对值后进行编码,因为边缘梯度值的正负号表示梯度两个对立的变化趋势,求绝对值也就忽视了梯度的变化趋势信息,而梯度的变化趋势有益于鉴别人脸特征的表达.针对LDP的不足提出了一种改进方案,直接利用局部邻域的原始边缘梯度值进行人脸特征提取.实验结果表明,改进方案的正确识别率高于原方案.
The local directional pattern（LDP） obtains the corresponding edge magnitude values by convolving the local 3×3 neighborhood with eight Kirsch masks.Then,the absolute edge magnitude values are sorted and the directions of three largest ones are encoded into an eight-bit binary number.The most shortcoming of LDP is the absolute operation of the edge magnitude values because the sign of the magnitude values means the two opposite trends（ascending or descending） of the gradient,which are very beneficial to the representation of the discriminant facial features.A novel approach based on the improved local directional pattern is presented for face recognition,which adopts the original edge gradient values to encode the facial features.The experimental results demonstrate that the proposed method is better than LDP.
出处
《宜宾学院学报》
2012年第6期77-80,共4页
Journal of Yibin University
基金
高等学校博士学科点专项科研基金(20090184110008)
四川省青年科技基金后续资助项目(09ZQ026-091)
关键词
局部邻域
边缘梯度
局部方向模式
人脸识别
鲁棒性
local neighborhood
edge gradient
local directional patterns
face recognition
robust
