基于FPCA和ReliefF算法的图像特征降维

针对传统图像特征降维方法计算量大、 无法去除冗余信息、 未考虑相关性等缺陷, 提出一种结合快速主成分分析(FPCA)和ReliefF算法的图像特征降维方法. 该方法先利用FPCAKG*6]算法对样本数据进行初次降维, 去除样本中的冗余信息；  再利用ReliefF算法计算样本特征的分类权重, 根据权重对特征进行组合优化. 在算法实现过程中, 采用递归排除策略, 进一步提升了算法特征寻优能力. 仿真实验表明, 利用本文算法优选出的图像特征, 可较好地提高聚类结果, 适合实际工程的应用.

Dimensionality Reduction for Image Feature Based on FPCA and ReliefF Algorithms

For the problems of a large amount of calculations, unremovable redundant information and unconsidered correlation in the traditional dimensionality reduction method for the image feature, a method based on the fast principal component analysis (FPCA) algorithm and the ReliefF algorithm was proposed. Firstly, the FPCA algorithm was used for the initial dimensionality reduction of the sample data to remove the redundant information; then the ReliefF algorithm was used to calculate the classification weights of the sample features which were used to perform optimized combination of features. In the algorithm implementation process, the recursive remove strategy was used to further enhance the ability of the algorithm to find the optimal characteristics. Simulation results show that the image features selected by the algorithm in this paper can better improve the clustering result, which is very suitable for practical engineering application.