摘要
为提高人脸识别的正确率,提出了一种改进的特征提取及分类算法。首先采用Contour-let变换对人脸图像进行多尺度分解,然后由低频子带和各尺度各方向的高频子带得到人脸的特征值,并将它们组合成多尺度特征向量,再应用多元回归分析方法进行人脸识别。由于多尺度特征向量不仅反映了整幅图像的全局特征,还反映了图像各种尺度下的边缘、纹理等奇异特征,因此具有更多的鉴别信息;多元回归分析则充分考虑了同一总体的各样本间的强线性关系。在ORL人脸库上的实验显示人脸识别率达97.78%,优于其他的方法。
In order to improve face recognition rate, an improved feature extraction and classification algorithm is proposed First face image is decomposed by using contourlet transform, and eigenvalues are obtained according to low-frequency sub-band and multi-scale, multi-direction higln frequency sub-bands. Then eigenvalues are combined as multi-scale feature vector, and classified by using multiple regression analysis. Multi-scale feature vector reflects not only the global features of the whole image, but also the singular characteristics such as edges and texture And therefore it has more identifying in{orrnatiorL Multiple regression analysis takes full account of the strong linear relationship between the various samples from the same population Experiments on the ORL face database show recognition rate of 97.78 %, better than the contrast methods.
出处
《光学与光电技术》
2012年第6期90-93,共4页
Optics & Optoelectronic Technology
关键词
人脸识别
多尺度特征向量
多元回归分析
特征提取
face recognition
multi scale feature vector
multiple regression analysis
feature extraction
