基于测度矩阵正则化的行人重识别算法

摘要

针对大多数的距离度量学习算法由于缺少可靠训练样本出现的过拟合现象,提出一种对测度矩阵的正则化方法.利用训练数据训练得到类间和类内协方差矩阵;利用倒数函数分别对两个协方差矩阵建立模型并进行正则化,获得正则化的测度矩阵;利用正则化后的测度矩阵对测试样本进行相似性度量.实验结果表明,该算法能够有效提高行人重识别精度.%Aiming at the over-fitting problem caused by less reliable training samples in most of current distance metric learning algorithms,a person re-identification based on regularized metric matrix algorithm was proposed.The training data were utilized to generate the within-class covariance matrix and between-class covariance matrix.The two covariance matrices were modeled and regularized in reciprocal function manner respectively so that regularized metric matrix was obtained.The similarity of test samples was calculated using the regularized metric matrix.The experiments verify that the proposed method improves the accuracy of the person re-identification efficiently.