Robust post-processing strategy for gait silhouette

An integrated silhouette with perfect appearance is helpful for gait recognition. However, the silhouettes often have holes or missing parts, because the commonly used motion detection and extraction methods are not always suitable for every case. A robust post-processing strategy is proposed here to refine the raw silhouettes. First, an individual silhouette model which represents the mutual characteristic of a certain sequence is trained to update the current sequence. Then a population silhouette model which represents the mutual characteristic of all sequences in the dataset is trained to update the sequences that need further refinement. The experiments on NLPR database show that the proposed algorithm is quite effective and helps the existing recognition method to achieve higher classification performance.