Boosting LDA with Regularization on MPCA Features for Gait Recognition

Abstract

In this paper, we present a boosted linear discriminant analysis (LDA) solution with regularization on features extracted by the multilinear principal component analysis (MPCA) for the gait recognition problem. This work is an extension of a recent LDA-based boosting approach and the MPCA is employed to project tensorial gait samples on a number of discriminative EigenTensorGaits (ETGs) to produce gait feature vectors for the base learners in boosting. This new scheme offers one more way to control the learner weakness while being very computationally efficient. Furthermore, the LDA learners are modified through regularization for protection against overfitting on the gallery set. Promising experimental results obtained on the Gait Challenge data sets indicate that the proposed algorithm is an efficient and effective solution consistently enhancing the gait recognition results on the seven probe sets by MPCA+LDA.