基于多目标进化封装的特征选择在人脸识别中的应用

为了在实施分类工作时将不相关的、多余的、具有噪声的特征从问题表示中去除，以降低复杂度并得到可接受的性能，提出了一种基于多目标进化封装的特征选择方法。首先利用染色体选择的特征重新参数化人脸图像从而获得主动形变模型特征集；然后通过多目标遗产算法进行特征选择，在最小化特征子集基数的同时最大化判别容量；最后结合提出的综合适应度函数及k-近邻分类器完成人脸的识别。在Essex人脸数据库上的实验验证了所提方法的有效性，实验结果表明，相比其它几种较为先进的方法，所提方法不仅降低了表示的维度，同时提高了分类性能。

Application of feature selection based on multi-objective evolutionary wrappers in face recognition

To perform the classification task with reduced complexity and acceptable performance by excluding irrelevant, redundant, or noisy from the problem representation, feature selection method based on multi-objective evolutionary wrappers is proposed. Firstly, features chosen by chromosome are used to parameterize face image again so as to get feature sets of active shape model. Then, multi-objective genetic algorithm is used to select features. Finally, proposed comprehensive fitness function and K nearest neighbor classifier are used to finish recognition. The efficiency of proposed method has been verified by experiments on Essex face database. Experimental results show that proposed method has improved the classification performance as well as reducing the representation dimensionality comparing with several advanced approaches.