View-Normalized Gait Recognition Based on Gait Frame Difference Entropy Image

Abstract

The difference of view in the gait image sequence can lead to the inconsistency of information contained in the different sequence of the same object, which affects the accuracy of feature extraction and increases the difficulty of recognition. Aiming at this problem, based on gait frame difference entropy image, this paper processes view normalization on the gait feature image using low rank optimization. Low rank optimization can keep the invariant part of the image to the maximum extent, reduce the influence of view change on feature. Finally, the nearest neighbor classification method is used to recognize. The experimental results show that the method of view normalization based on gait frame difference entropy image improves the recognition rate under cross view to a certain extent, and has some robustness to walking state and clothing change.