针对LDP利用Kirsch算子计算8方向的边缘响应值并排序,特征提取速度慢的问题,提出了一种改进的分解局部方向模式DLDP(divided local directional pattern)特征提取方法。将Kirsch算子的8个方向掩模分成2个子方向掩模再分别计算边缘响应值...针对LDP利用Kirsch算子计算8方向的边缘响应值并排序,特征提取速度慢的问题,提出了一种改进的分解局部方向模式DLDP(divided local directional pattern)特征提取方法。将Kirsch算子的8个方向掩模分成2个子方向掩模再分别计算边缘响应值,获得2个编码(DLDP1和DLDP2),级联两个编码的直方图得到表情特征DLDP。然后利用主成分分析法(PCA,principal component analysis)降维处理。最后用支持向量机进行表情识别,在JAFFE数据库上的实验表明,本文方法与近几年效果较好的特征提取算法相比,不仅缩短了特征提取的运算时间,而且提高了识别率。展开更多
The local directional pattern (LDP) is unsusceptible to random noise which is widely used in texture extraction of face region. LDP cannot encode the central pixel thus the important information will be lost. Thus a...The local directional pattern (LDP) is unsusceptible to random noise which is widely used in texture extraction of face region. LDP cannot encode the central pixel thus the important information will be lost. Thus a new feature descriptor called extended local directional pattern (ELDP) is proposed for face extraction. First, the mean value of the eight directional edge response values and the gray value of center pixel are calculated. Second, the mean value is taken as the threshold. Then, the expression image is encoded using nine encoded values. In order to reduce redundant information and get more effective information, the Gabor filter is used to obtain the multi- direction Gabor magnitude maps (GMMs) , and then the ELDP is used to encode the GMMs. Finally, support vector machine (SVM) is applied to classify and recognize facial expression. The experimental results show that the feature dimensions is greatly reduced and the rate of facial expression recognition is improved.展开更多
基金supported by the Science and Technology Commission Project of Chongqing ( cstc2015jcyj BX0066)
文摘The local directional pattern (LDP) is unsusceptible to random noise which is widely used in texture extraction of face region. LDP cannot encode the central pixel thus the important information will be lost. Thus a new feature descriptor called extended local directional pattern (ELDP) is proposed for face extraction. First, the mean value of the eight directional edge response values and the gray value of center pixel are calculated. Second, the mean value is taken as the threshold. Then, the expression image is encoded using nine encoded values. In order to reduce redundant information and get more effective information, the Gabor filter is used to obtain the multi- direction Gabor magnitude maps (GMMs) , and then the ELDP is used to encode the GMMs. Finally, support vector machine (SVM) is applied to classify and recognize facial expression. The experimental results show that the feature dimensions is greatly reduced and the rate of facial expression recognition is improved.