Part-based gait identification using fusion technique

Appearance based gait recognition is largely affected by various covariate factors. This problem can be reduced significantly by dividing the gait silhouette into parts. Some works have already been done related to part-based gait identification, where some parts are considered as redundant. Because every part carries discriminating information more or less, we propose a way to combine all the parts by assigning weights to each of them. Weights are calculated using the training set. In this work we have used Gait Energy Image (GEI), Discrete Fourier Transformation (DFT) Gait Image and Gait Entropy Image (GEnI) representations of binary gait silhouettes. The results of weighted fusion of parts on both DFT and GEnI representations give outstanding performance compared to previous approaches on gait recognition.