Multilinear mean component analysis for gait recognition

In this paper multilinear mean component analysis (MMCA) is introduced as a new algorithm for gait recognition. Compared with traditional PCA and MPCA, the new method MMCA is based on dimensionality reduction by preserving the squared length, and implicitly also the direction of the mean vector of the each mode´s original input. The solution is not necessarily corresponding to the top eigenvalues. MMCA improved the clustering results and reduced the small sample size (SSS) problem and has great convergence. MMCA as a feature extraction tool provides stable recognition rates and the MMCA-based approaches we proposed achieves better performance for gait recognition based on the University of South Florida (USF) HumanID Database.