摘要
针对经典PCA算法在人脸特征提取上的局限性,提出了一种基于克隆选择算法的特征选择方法.克隆选择算法的收敛速度快,具有较强的全局搜索能力,可以快速搜索到最有利于分类的特征空间;因此利用克隆选择算法对PCA变换后的特征向量进行选择,可以有效避免PCA只选择人脸轮廓信息,而忽略细节信息的不足,在人脸识别中取得了较好的效果.通过对ORL和Yale人脸库的仿真实验表明,该方法无论在识别率、降维效果还是稳定性方面,性能均优于遗传算法,不但有效降低了特征向量维数,还将人脸识别率提高到91.5%,因此研究该算法具有很强的现实意义.
Limitations of classic PCA algorithm in the face feature extraction were illustrated, and a new feature selection method based on colonial selection algorithm was proposed. Colonial selection algorithm enjoys fast convergence and strong global searching capability. It can quickly find out character space that is most beneficial to classification, so colonial selection algorithm is used in feature vector selection after the transformation of PCA, so as to effectively avoid choosing only the outline of face and neglecting the details. Simulation experiments on ORL and Yale face database show that in terms of the recognition rate, the effect of reduced dimension and stability, this new method is superior to genetic algorithm. Therefore, the study of this method is of practical significance.
出处
《应用科技》
CAS
2009年第3期11-14,共4页
Applied Science and Technology
基金
国家自然科学基金资助项目(60672034)
