中文摘要：

针对手形特征识别中，由于特征间高相关性产生冗余而降低识别性能的问题，提出利用信息增益和相关系数分别对特征的分类区分度和相关性进行评价，并经过综合分析对手形特征进行优化选择。该方法能够保留分类中起关键作用的特征，并同时去除高相关性的冗余特征量。为了证明该方法的有效性和准确性，采用浮动搜索的方法，以识别率为评价函数确定特征优化组合。实验结果表明，优化后6个特征组成向量的识别率达到96．24％，比全部9个特征组成的特征向量提高了0．43％，同时由于特征数目的减少也降低了运算时间。该方法可以避免常用的搜索性选择方法的复杂性，并有效去除手形识别中低区分度和高冗余的特征，有利于简化算法并与其他特征进行融合使用。

英文摘要：

Aiming at the problem in hand-shape feature recognition that the high correlation among the features causes redundancy and the identification performance decreases, this paper proposes a feature selection method, which uses the information gain and correlation coefficient to evaluate the class distinctive degree and correlation of the fea- tures, respectively;and through comprehensively analyzing, the hand-shape features are selected optimally. The meth- od can reserve the features that play a key role in classification and remove the redundant features with high correla- tion at the same time. In order to prove the validity and accuracy of the method, the sequential forward floating selec- tion algorithm is adopted to generate subsets, and the recognition rate is taken as evaluation function to determine the optimized feature combination. The experiment results show that using the feature vector composed of optimized 6 fea- tures, the recognition rate reaches 96.24% , which is 0.43% higher than that using the feature vector composed of to- tal 9 features. At the same time, because the number of features decreases the operation time is reduced. This method can avoid the complexity of common searching selection methods and effectively remove the low distinctive degree and high-redundancy features in hand shape identification ,which benefits algorithm simplification, and can be fused with other features.