CAJ | 학술논문

针对 AdaBoost算法忽略弱分类器之间的相关性导致强分类器的集成性能降低的缺陷，提出一种改进的 AdaBoost人脸检测算法。通过加入差异性度量进行相关性判定，并根据判定值剔除相似特征，以有效增加分类器的多样性。实验结果表明，相同条件下，改进的算法提高了人脸检测率，降低了错误检测数。
침대 AdaBoost산법홀략약분류기지간적상관성도치강분류기적집성성능강저적결함，제출일충개진적 AdaBoost인검검측산법。통과가입차이성도량진행상관성판정，병근거판정치척제상사특정，이유효증가분류기적다양성。실험결과표명，상동조건하，개진적산법제고료인검검측솔，강저료착오검측수。
Aiming at the problem of traditional AdaBoost algorithm ignoring the correlation between weak classifiers,which reduces the ensemble of the AdaBoost algorithm,an improved AdaBoost face detection algorithm is proposed.The algorithm uses the disagreement measure as the degrees of diversity between classifiers and eliminates similar features according to de-cision value,so as to effectively improve the weak classifier diversity.Experimental results show that the algorithm im-proves the face detection rate and reduces false alarm rate in the same conditions.