基于多类在线Boosting的图像识别算法
Image Recognition Algorithm Based on Multiclass Online Boosting
-
摘要: 针对在线Boosting算法难以在多类图像识别中使用的问题,提出了一种基于错误纠正输出编码(ECOC)的多类在线Boosting算法.该算法在计算弱分类器的错误率时借鉴ECOC的思想,引入了一个类别标签映射函数;然后给出了在该映射函数下训练样本的权重及弱分类器的权重的计算与更新方法.通过在不同数据库上的对比实验,验证了文中算法是快速有效的,且具有较强的鲁棒性.Abstract: Aiming at the difficulty of applying online Boosting to multiclass image recognition,a new online Boosting algorithm based on error-correcting output code(ECOC) is presented.In this algorithm,a function for class label mapping is used when calculating the error rates of weak classifiers.Furthermore,the method to update the weights of training samples and weak classifiers are both given.The experiment results on different database show this algorithm is accurate and robust.