摘要
提出了一种新的Adaboost快速训练方法,解决了基于Adaboost的人脸检测算法中结构复杂、训练非常耗时的问题.新方法从两方面提高训练速度:直接求解训练中Adaboost目标函数;在直接求解算法基础上,使用了双阈值简单分类器构造强分类器的Adaboost检测器结构.
Recently the human face detection system based on Adaboost is successfully used in application areas because of its high speed and accepted detection rates, but building this system is very complex and its training time is extremely long. Numerous weaker classifiers need to be updated in the Adaboost during the training stage. A new fast training algorithm for Adaboost is proposed to solve this problem. Two methods are adopted to accelerate the training:(1) A method to directly solve the parameters of single weaker classifier is proposed, making the training speed is higher than probability method about 20 times and higher than artificial neural network thousands of times; (2) A double threshold decision for single weaker classifier is introduced, and the number of weaker classifiers in the Adaboost system is reduced, which simplifies the structure of the detection system. Based on the simplified detector, both the training time and the detecting time can be reduced.
出处
《复旦学报（自然科学版）》
CAS
CSCD
北大核心
2004年第1期27-33,共7页
Journal of Fudan University：Natural Science
基金
国家自然科学基金资助项目(60171037)
