摘要： 为解决基于Adaboost算法的人脸检测训练耗时的问题，提出一种Adaboost快速训练算法。基于原算法，在训练中使用序列化表格选取弱特征，在一轮训练结束后不进行样本权值更新，直接在已选分类器的基础上利用直方图统计的方法进行下一轮训练。实验证明该算法有较高的训练效率。

关键词: 人脸检测, 分类器, 训练算法

Abstract: In order to solve the problem of long training time in face detection with Adaboost algorithm, this paper presents a fast Adaboost training algorithm. Based on original Adaboost, it uses the serialization table to select weaker classifier in training process. After a round of training, it does not update the sample weights but does the next round training by using histogram statistics, which is based on the selected classifier. Experimental result show that the algorithm gains higher training efficiency.

Key words: face detection, classifier, training algorithm

中图分类号: 

• TP391.41