CAJ | 학술논문

提出了一种类似于聚类分析的融合方法，它通过分析样本在特征空间的分布，来估计分类器分类结果的可靠性，并根据各个样本的具体情况自适应地为各分类器赋予权值，从数据融合的层次上来说，这是一种介于特征级和决策级的融合方法.
제출료일충유사우취류분석적융합방법，타통과분석양본재특정공간적분포，래고계분류기분류결과적가고성，병근거각개양본적구체정황자괄응지위각분류기부여권치，종수거융합적층차상래설，저시일충개우특정급화결책급적융합방법.
Combining information from multiple classifiers can improve the perfor mance of pattern recognition systems. However, the traditional methods always as sign fixed weights to the classifiers according to their classification performa nces without considering the sample itself. A clusterirng analogy fusion method is proposed in this paper, which estimates the reliability of each classifier by analyzing the distribution of samples, and adaptively assigns weights to classi fiers based on the reliability estimation. This method can be seen as a method l ying between feature level and decision level.