Abstract:A feature representation method based on multi-layer local moment invariants for gait analysis and recognition applications is developed. The method includes following steps: first, silhouette extraction is performed for each image sequence. Secondly, the gait cycle is detected by a histogram-based approach. Thirdly, a scalarvalued CGHI(Colored Gait History Images) is proposed to describe how human walking is evolved. To improve the recognition rate, the CGHI is decomposed into a sequnce of muti-layer rectangle windows. The moment invariants from the window are used as the gait features and finally used to recognize gait. The correct classification rate of 87.2% is achieved on Soton database, which show that the method outperforms the exist methods.