改进的SVM分类算法在人脸年龄估计中的应用

Facial Age Estimation Based on Improved SVM Classification

支持向量机的训练需要求解一个带约束的二次规划问题,但在数据规模很大情况下,经典训练方法将变得很困难。提出一种基于改进的混合蛙跳算法的SVM训练算法。针对混合蛙跳算法搜索速度慢且容易陷入局部极值的缺陷,将模拟退火思想引入到混合蛙跳算法中,提出一种改进的混合蛙跳算法,并将其应用到人脸年龄估计中去。另外使用核主成分分析算法、Gabor小波变换以及局域二值变换来提取人脸的特征,将这3种特征分别特征层和决策层融合后,得到更为适合人脸年龄的特征向量。实验结果表明,使用该算法得到的人脸年龄段分类的分类准确率相对较高。

Since training SVM requires solving a restrained quadratic programming problem which becomes difficult for large datasets,a improved Shuffled Frog Leaping Algorithm(SFLA)is proposed as an alternative to current algorithm.In order to overcome the defects of SFLA such as slow searching speed in evolution and local minimum,an improved algorithm in which the mechanism of Simulated Annealing(SA)is involved into basic SFLA is put forward.And it is applied into the facial age estimation.Besides the kernel Principal Component Analysis(PCA),Gabor wavelet transform as well as the LBP arithmetic are used as the feature extraction methods.Fusing these three feature extraction method in feature level and decision-making level,a more suitable method for extracting facial aging feature is obtained.The test results indicate that the algorithm enhances the convergence velocity outstandingly and averting the local extreme values effectively,and it is effective and feasible for SVM training,besides,the classification accuracy of age group is relatively higher.