摘要
AdaBoost是一种构建准确分类器的学习算法,但其训练样本时间长制约了发展.本文对训练算法进行改进,通过直方图将弱分类器学习训练从循环中提取出来,以缩短训练时间;且在人脸检测方面,变标准步长为动态步长,有效地避免冗余计算,提高检测速度.实验表明通过两方面的改进,提高了检测速度,因此在实时性要求较高的应用领域有现实意义.
AdaBoost is a learning algorithm to build an accurate classifier.But its development is restricted,because the time of training sample is too long.So this paper intends to focus on its improvement.On the basis of the histogram,the weak classifier trainingwas selected and extracted from the circulation,which greatly reduced the training time.Moreover in the face detection,the standard step was replaced by dynamic step,and it effectively avoided the redundant computation in the detection process and quickened the detection.The experiments showed thatwith the improvements of the two aspects,this algorithm had practical significance in the application field of higher real-time requirements.
出处
《湘南学院学报》
2013年第5期23-26,79,共5页
Journal of Xiangnan University
基金
湘南学院质量工程项目(湘南学院校发[2013]157号)
关键词
人脸检测
弱分类器
动态步长
face detection
weak classifier
dynamic step
