Sparse error gait image: a new representation for gait recognition

Abstract

The performance of a gait recognition system is very much related to the usage of efficient feature representation and recognition modules. The first extracts features from an input image sequence to represent a user's distinctive gait pattern. The recognition module then compares the features of a probe user with those registered in the gallery database. This paper presents a novel gait feature representation, called Sparse Error Gait Image (SEGI), derived from the application of Robust Principal Component Analysis (RPCA) to Gait Energy Images (GEI). GEIs obtained from the same user at different instants always present some differences. Applying RPCA results in low-rank and sparse error components, the former capturing the commonalities and encompassing the small differences between input GEIs, while the larger differences are captured by the sparse error component. The proposed SEGI representation exploits the latter for recognition purposes. This paper also proposes two simple approaches for the recognition module, to exploit the SEGI, based on the computation of a Euclidean norm or the Euclidean distance. Using these simple recognition methods and the proposed SEGI representation gait recognition, results equivalent to the state-of-the-art are obtained.