基于敏感度分析的图像特征值选取

特征选择在数据挖掘、图像识别等诸多方面有着广泛的应用,其目的是找出那些最有效的特征,即把特征空间从高维压缩到低维。对于图像识别系统而言,为了保证识别性能需要从图像中提取大量的信息,往往使得训练集数量相对特征向量的维数显得较少。引入敏感度分析作为标准实现图像特征值的选取。实验表明:利用敏感度分析选取的特征值对BP神经网络进行训练避免了网络的过拟合问题,提高了网络的识别率,同时大大降低了网络的训练时间,提高了网络识别效率。

Image Feature Selection Based on Sensitivity Analysis

Feature selection has been applied to several fields like data mining,image recognization and so on,which purpose is selecting these most effective features from feature vector and reducing the dimension of vector.For image recognization system,in order to make sure the system performance we have to get a lot of information from image,which may make the training samples is not enough in contrast to the dimension of vector.In this paper,sensitivity analysis is applied to select features.It has been proved that using features selected by sensitivity analysis to train BP neural net can avoid net overfitting,improve the performance of the system,greatly reduce the time consuming and make recognization system more effectively.