摘要
提出了一种改进的基于多区块局部二值模式(MB-LBP)特征的人脸检测算法。算法针对AdaBoost算法训练过程中出现的权值分布扭曲的现象,对样本权值的更新规则进行了调整。实验结果表明,该方法有效地缩短了训练时间,而且避免了权值扭曲的现象。算法在保证检测率的同时降低了误检率。
In this paper,an improved face detection method called MBLBP-AdaBoost algorithm which based on Multiblock Local Binary Patterns (MB-LBP)features is presented. And aimed at the phenomenon of weights distortions in training process of AdaBoost algorithm,the algorithm of sample weights updated rules have been adjusted. Ours experimental results show that the new method can effectively shorten the training time, and avoids the phenomenon of weights distortions. The algorithm reduee false alarm rate while holding a high detection rate when testing in the CMU +MIT databases.
出处
《科技通报》
北大核心
2011年第5期652-656,共5页
Bulletin of Science and Technology
基金
国家自然科学基金资助项目(61005008
90820009)