摘要
提出增强型Haar-Like特征和基于双阈值的弱分类器快速训练方法,然后在此基础上提出了继承型Ad-aBoost算法,后层分类器的训练是建立在前层分类器基础之上。该方法具有很快的训练速度,弥补了AdaBoost算法训练速度慢,使得弱分类器数目大大减少,还使得总体检测器的性能得以加强。在实验过程中,首先使用MIT-CBCL库在同样的训练样本和测试样本条件下对几种方法进行了实验,结果表明该方法在训练速度、测试精度及检测时间等方面都优于相应的方法,最后在MIT+CMU人脸测试库进一步验证算法的有效性。
A fast face detection method is presented based on the heritance AdaBoost algorithm. The method remedy defects of the slow training speed, the reduced weak classier of curent AdaBoost. Firstly, an enhanced Haar-Like features and a fast training method of the weak classier with double-threshold are developed. Then, a heritance AdaBoost algorithm used to train classier by training results of prior classier is proposed. The algorithm reduces training time and the number of the weak classier, and enhances the performance of Cascade detector. In the experimental process, several methods are used to MIT-CBCL dataset on the same training samples and experimental samples. Experimental results show that the training speed, precision and detection time of the method are superior to the corresponding methods. Finally, experimental results on MIT＋CMU dataset demonstrate that the algorithm is efficient.
出处
《数据采集与处理》
CSCD
北大核心
2008年第3期306-310,共5页
Journal of Data Acquisition and Processing
基金
国家自然科学基金(60632050(重点项目),60472060,60473039)资助项目
江苏省高技术研究基金(BG2005008)资助项目
