基于直方图的mass函数构造方法

A Mass Function Construction Method Based on Histogram

针对多特征融合的模式识别问题,给出了一种利用样本特征分布的直方图构造mass函数的方法。首先做出样本特征的直方图;在特征直方图的重叠区域,特征的不确定性较大;在特征直方图的非重叠区域,特征的确定性较大。然后,对于一个新的对象,若它的某一特征落入直方图的重叠区,由该特征构造的mass函数有较大的不确定性;若该特征落入直方图的非重叠区,则由该特征构成的mass函数确定性较大。把不同特征的mass函数进行融合得到最终的融合结果。对鸢尾属植物进行分类实验的正确率达到96.64%,实验结果表明了该方法的有效性。

A method based on the histogram of the sample feature distribution is presented to construct the mass function,for the problem of pattern recognition using the multi-feature fusion.Firstly,the sam-ple feature distribution is established.In the overlapping area of the histogram,the feature is uncertain;while in the no-overlapping area of the histogram the feature is determinate.Then,for a new object,if one of its features falls into the overlapping region of the histogram,the mass function constructed by this feature has a larger uncertainty;if the feature falls into non-overlapping region of the histogram,the mass function constructed by this feature has a greater certainty.The mass functions of different features are fused to get the fusion result.The correct ratio of the iris-plant classify experiment is 96.64%,and the result shows that this method is feasible.