基于Adaboost的人脸检测中关于级联分类器设计和训练问题的探究

摘要

本文通过基于Adaboost实现的人脸检测实验，总结了在此方法的程序设计中需要关注的几个问题，并根据实际的实验数据给出了针对这些问题的解决方法。在去光照问题上，介绍了Viola和Jones的局部归一化解决方法。在级联分类器的设计和训练中，介绍了级联结构的设计，以及在每一级的训练中训练数据的设计及训练方法。%In this paper ,based on the practice of face detecting using Adaboost algorithm ,several problems are listed that need to be focused on in programming .In order to eliminate the impact of light, Viola and Jones presented a local normalization method which is simple and effective .The design of cascade classifiers is showed next. In the phase of planning the structure of training data ,how the positive and negative samples are organized and taken from is proposed .