基于决策树的模糊序贯最小优化分类器的人脸识别

序贯最小优化算法是一种SVM s(Support VectorM ach ines)训练算法,该算法将一个大型QP(Quadratic Programm ing)问题分解为一系列最小规模的QP子问题,从而避免了多样本情形下的数值解不稳定及耗时问题,同时也不需要大的矩阵存储空间。本文在模糊支持向量机的基础上,提出了基于决策树的模糊序贯最小优化算法并对它进行了分析和研究,在对人脸图像进行独立成分分析后,用该算法进行多类人脸识别。通过在ORL人脸库上的实验结果表明,在样本类别较少的条件下,该算法可以取得较好的效果。

Face Recognition Based on Fuzzy Sequential Minimal Optimization and Decision Tree

SMO(Sequential Minimal Optimization) is a simple algorithm that can solve the quadratic programming problem of SVM(Support Vector Machine) quickly without any extra matrix storage and numerical optimization steps.SMO can decompose the overall quadratic programming problem into sub-problems by using Osuna's theorem to ensure convergence.DT-FSMO based on the combination of SMO and the fuzzy support vector machine based on decision tree(DT-FSVM) is suggested in which the features are extracted from the original face images through ICA.The comparison is given for the DT-FSMO and DT-FSVM.The experimental results conducted on ORL face images show that a better effect with this method can be obtained provided that the number of classes is relatively small.