【摘要】 利用矩阵特征向量分解,提出一种可最优化计算特征规模的互信息特征提取方法.首先,论述了高斯分布假设下的该互信息判据的类可分特性,并证明了现有典型算法都是本算法的特例;然后,在给出该互信息判据严格的数学意义基础上,提出了基于矩阵特征向量分解计算最优化特征规模算法;最后,通过实际数据验证了该方法的有效性.更多还原

【Abstract】 By using matrix eigenvalue/eigenvector decomposition,an optimization calculation of feature scale is proposed for mutual information measure feature extraction technique.The novel technique based on Gaussian distribution enjoys a good class-separability property with the mutual information.It is proved that the existing algorithm is the specifical example.Then based on the strict mathematic significance of mutual information measure,the method for optimization calculation of the feature scale based on matrix eigenvalue/eigenvector decomposition is given.Finally,real-world data sets demonstrat the effectiveness of the method.更多还原