基于改进最小最大准则的分类器设计研究

主要探讨先验概率未知情况下的分类器设计问题。为了解决传统的最小最大分类器性能有限的不足,提出了基于分段线性化的分类器设计方法。方法不仅是最小最大准则的改进,而且也是最优贝叶斯分类器的更好近似。通过说话人识别的应用,验证了所提出算法的有效性。

Classifier Design Research towards Improved Minimax Approach

The problem of designing a classifier when prior probabilities are not precisely known is discussed in this paper. In order to overcome the shortcomings of a minimax classifier which gives a relatively poor performance at most of possible values of prior probabilities, we present a novel classifier design method based on piecewise linear approximation. The proposed switched classifier is not only an improvement of the minimax classifier but also a good approximation to the Bayes classifier. To justify the proposed method, we also addressed the problem of speaker recognition in a real application by using the presented classifier.