In this paper, we propose a frontal face detection method based on Support Vector Machine (SVM) and Histogram of Oriented Gradients (HOG).Support Vector Machine selects support vectors according to the HOG feature and uses these support vectors to build the classiifer.The training and testing positive samples are all selected from the CMU PIE multi-pose and multi-illumination face database,the negative samples are selected from the Internet, the sample size is normalized to 20 × 20 pixels. The classifier of the detection system is a support vector machine, whose kernel function is linear. The feature we choose is ifrstly raised by Navneet Dalal and Bill Triggs in pedestrian detection issues,by selecting the appropriate parameters, we obtain a 384-dimensional feature, the cell size of the feature(Histogram of Oriented Gradients) is 4×4 pixels, each block contains four cells, and each cell contains six bins. The classiifer we trained has the detection rate of 92%on the test set and the false alarm rate is also low .By comparing the result of our method with the result of the face detection method based on adaboost of opencv, the result shows that our face detection system is quite good. In the CMU+MIT frontal face test set this method also achieved good results. Experimental results show that the proposed method in face detection problem is relatively effective.%本文提出了一种基于支持向量机和方向梯度直方图的正面人脸检测方法。支持向量机通过学习方向梯度直方图特征来选取支持向量，然后根据这些支持向量构建最优分类面。实验使用的训练样本和测试样本从CMU的PIE多姿态和多光照人脸数据库中选取，样本大小被标准化为20×20像素。检测系统选用的分类器是支持向量机，其核函数是线性的。选用的特征是Navneet Dalal和Bill Triggs在行人检测问题上提出的方向梯度直方图。训练好的分类器在测试集合上的检出率为92%。在CMU+MIT正面人脸测试集合上也取得了较好的结果。实验结果表明，本文提出的方法在人脸检测问题上是比较有效的。
展开▼
