一种新的多分类器融合方法

On a New Multiple Classifiers Ensemble Method

提出了一种新的基于多分类器联合的模式识别方法,该方法引入图像处理领域的对比度概念,对多分类器系统中不同基本分类器的融合权重动态赋值,并通过融合权重联合所有基本分类器决策信息,同时构建一个临时的全局分类器来做出模式识别任务的融合决策输出.试验表明该方法在模式识别性能上能够获得较好的性能.

This paper proposes a new pattern recognition method based on multiple classifiers ensemble. The new method calculates weights of each base classifier in multiple classifier system with the contrast of base classifier’s decision outputs.Then,a global classifier can be built by fusing all base classifiers’deci-sion outputs.And it is used to make final decision.Experiment results show this new method can get higher accuracy than single classifier.