一种改进的强分类器优化算法

在Gentle Adaboost算法中,强分类器由弱分类器线性组成,但这种组合并不能保证强分类器是最优的.因此,提出一种基于分类结果的优化方法.算法模拟弱分类器系数增加或减少对分类结果的影响,选择最利于分类的系数调整方向,设计了优化机制.在MIT人脸数据库上的仿真实验表明,分类器优化后,检测精度得到了提高.

Improved strong classifier optimization algorithm

Strong classifier is a linear combination of weak classifiers in Gentle Adaboost algorithm, but this kind of combination can not ensure the strong classifier optimal. An optimization method based on classification result was proposed. The algorithm simulated the effect of coeffcient adjustment on the classifier result, chose the best direction of the classifier result and designed the optimization mechanism. The proposed method can optimize the classifier and the simulation results on the MIT human face database show the detection accuracy has been enhanced.