为了实现复杂环境下的人脸特征有效表达，提出一种改进的梯度方向直方图（ＨＯＧ）人脸识别方法．首先以人脸图像网格作为采样窗口并在其上提取 ＨＯＧ特征；然后将所有网格 ＨＯＧ特征向量进行组合，实现整个人脸特

征表达；最后采用最近邻分类器进行识别．另外，比较了该方法与Ｇａｂｏｒ小波和局部二值模式（ＬＢＰ）２种著名的人脸

局部特征表示方法的优劣．实验结果表明，在调优的 ＨＯＧ参数下，在具有光照和时间环境等复杂变化的ＦＥＲＥＴ人

脸库中，较少维数的 ＨＯＧ特征比ＬＢＰ特征有更好的表现，而且 ＨＯＧ特征提取时间和特征向量维数比Ｇａｂｏｒ小波方法更具有优势-In order to achieve facial features in complex environments valid expression, an improved gradient direction histogram (HOG) face recognition method. Firstly face image and extract the grid as a sample window HOG features on it　then all mesh HOG feature vector combination, realize the whole people express facial feature　Finally, nearest neighbor classifier to identify. In addition, the comparison of the method with Gabor wavelet and local binary pattern (LBP) 2 famous facial features indicate the quality of the local approach. Experimental results show that HOG parameter tuning in FERET face　with complex changes in the environment of light and time, the characteristic dimension of less than HOG LBP features better performance, and feature extraction time and HOG dimension of feature vectors have an advantage over Gabor wavelet method