一种改进的基于ICA特征子空间的目标识别方法

介绍了独立分量分析（ICA）的基本原理和算法，并提出了基于独立分量分析的特征子空间的目标识别方法。该方法首先利用快速独立分量分析（FastICA）算法对训练集目标图像进行ICA分解，据此建立特征子空间，然后根据待识别图像在特征子空间的投影系数进行判别。本文的改进在于根据类内类间距离比值最小化准则进行最有利于分类的特征的优化选择。实验结果显示，和传统方法相比，改进的方法能有效提高识别的准确率和效率。

Improved Method for Object Recognition Based on Subspace of Features Using Optimal Selection of Features

In this paper,the principle and the algorithm of Independent Component Analysis(ICA) are described.Based on the subspace of features using independent components,a method for the recognition of objects is proposed.Firstly,the FastICA algorithm is perfomed on the training group of images for the objects. Secondly,the subspace of features is constructed according to the extracted independent components.Lastly,the recognition of objects is done according to the projection coefficients of images to the subspace of features.In this improved method,the minimum rule of the ratio between the distances of classes is applied to select the optimal features favourable for the recognition.Demonstration indicates that the improved method has reached a higher recognition rate with less time cost compared with the traditional methods.