Feature extraction based on a fuzzy complementary criterion for gait recognition using GRF signals

A novel wavelet-based feature extraction approach is introduced in this paper for subject recognition utilizing ground reaction force (GRF) measurements. A wavelet-packet (WP) decomposition scheme is firstly proposed to recognize the discriminating frequency subbands and subsequently an efficient feature selection (FS) method is applied on the selected WP bands providing a compact set of powerful and complementary features. Our approach relies on a non-global fuzzy set-based criterion to assess the significance of every subband or feature. This local evaluation measure with respect to patterns is implemented by a fuzzy partition vector (FPV) constructed by invoking a fuzzy class allocation scheme that assigns membership grades to every class. The FS is driven by a fuzzy complementary criterion (FuzCoC) that acts upon the feature FPVs, handling simultaneously both the discrimination power and the redundancy between the features. To demonstrate the performance capabilities of our approach an extensive experimental setup is designed with tasks of increasing difficulty.