The present invention relates to a face detection method, and relates to a new multi-view face detection method (MVFD) based on an improved real AdaBoost algorithm, wherein an improved real AdaBoost algorithm for a trained Haar feature is provided, Design a divided sample. The time complexity of the weak classifier is 0 (M * N), the time complexity of the strong classifier is 0 (T * M * N), and the training speed is 0 (N) . Based on the processing of our Haar feature for multi-view face recognition (MVFD), only 12 category sorters are needed to generate ROPs and change the angle of the RIP. The algorithm devised above is characterized by greatly reducing training complexity. To use the real AdaBoost algorithm, the LUT weak classifier is used to enhance the Haar feature. Instead of each trained pose measurer, we use the confidence of the strong classifier defined by the AdaBoost algorithm. The first four of the sixteen layers of the cascaded classifier are used for pose estimation, and no extra computation for pose estimation is necessary. Algorithm tests based on CMU and MIT face data bases were performed and the experimental results show that the convergence performance of the improved real AdaBoost algorithm is improved over the previous real AdaBoost algorithm. In addition, the multi-view face detection system designed in the present invention provides a high detection rate, excellent timeliness, and a high detection rate.
展开▼
机译:本发明涉及一种面部检测方法,并且涉及一种基于改进的真实AdaBoost算法的新的多视角面部检测方法(MVFD),其中提供了一种用于训练Haar特征的改进的真实AdaBoost算法,设计了划分样本。弱分类器的时间复杂度为0(M * N),强分类器的时间复杂度为0(T * M * N),训练速度为0(N)。基于我们用于多视图面部识别(MVFD)的Haar功能的处理,仅需要12个类别分类器即可生成ROP并更改RIP的角度。上面设计的算法的特征在于大大降低了训练复杂度。为了使用真实的AdaBoost算法,使用LUT弱分类器来增强Haar功能。代替每个受过训练的姿势测量器,我们使用AdaBoost算法定义的强分类器的置信度。级联分类器的十六个层中的前四个用于姿势估计,并且不需要额外的计算来进行姿势估计。进行了基于CMU和MIT人脸数据库的算法测试,实验结果表明,改进的真实AdaBoost算法的收敛性能比以前的真实AdaBoost算法有所提高。另外,本发明设计的多视点面部检测系统提供了高检测率,优异的时效性和高检测率。
展开▼