Modified Marginal Fisher Analysis for Gait Image Dimensionality Reduction and Classification

Abstract

Gait is a kind of biometric feature to identify a walking person at a distance. As an important biometric feature, human gait has great potential in video-surveillance-based applications, which aims to recognize people by a sequence of walking images. Compared with other biometric feature identifications such as face, fingerprint or iris, in medium to long distance security and surveillance applications in public space, the most important advantage of gait identification is that it can be done at a distance. As gait images are complex, time-varying, high-dimensionality and nonlinear data, many classical pattern recognition methods cannot be applied to gait recognition directly. The main problem in gait recognition asks is dimensionality reduction. Marginal Fisher analysis (MFA) is an efficient and robust dimensionality reduction algorithm. However, MFA does not take the data distribution into consideration. Based on original MFA, a modified MFA is proposed for gait recognition. Firstly, the discriminant classification information is computed to guide the procedure of extracting intrinsic low-dimensional features and provides a linear projection matrix, and then both the between-class and the within-class scatter matrices are redefined by the classification probability. Secondly, through maximizing the between-class scatter and minimizing the within-class scatter simultaneously, a projection matrix can be computed and the high-dimensional data are projected to the low-dimensional feature space. The experimental results on gait database demonstrate the effectiveness of the proposed method.