步态特征中非线性相关信息的表示方法研究

Gait Representation with the Non-linearity Correlation Information

针对步态特征提取时非线性相关信息的表示问题,基于典型相关分析,结合局部化思想,提出局部判别典型相关分析方法进行步态特征表达并进行步态识别,提取的步态特征使得同类样本之间的相关度最大化,而不同类样本之间的相关度最小化.实验结果表明该算法有效地将非线性相关信息结合到步态特征后,识别性能相对典型相关分析明显提高。

To deal with how to characterize the gait with non-linearity correlation information, combined the Canonical Correlation Analysis with the locality method, the locality Discriminant CCA is used to the gait recognition. The features extracted by LDCCA maximized the local within-class correlations, and minimized the local between-class correlation. It is so important that the performance of the pattern recognition can be improved. The experimental results demonstrate that since the non-linearity correlation information could be used effectually, the LDCCA outperform CCA to get the encouraging performance, as well as the more actual gait feature.