A score level fusion framework for gait-based human recognition

Three different contour features are fused for gait recognition through a score level information fusion framework. The first contour feature is procrustes mean shape (PMS) which is the compact representation of gait sequence. The other two features are proposed based on PMS. One is shape context features which utilize the shape context descriptor to depict the global distribution of sample points on PMS. The other is a local discriminative gait feature called tangent angle features which concentrate on the local characteristic of adjacent points on PMS. At last, those three features are fused at matching score level with five different rules. Large amount of experiments on CASIA and SOTON datasets show the proposed new contour features are more effective than the original one, and also demonstrate that the proposed fusion algorithm outperforms other algorithms.