Optimizing feature selection using laplace similarity in occluded human motion recognition

In this paper, feature selection is used to allow the identification of critical attributes before the reconstruction of occluded data in 3d human motion classification. This work presents Fuzzy Laplace Similarity: a new fuzzy similarity relation used in the context of fuzzy rough feature selection. The measure of similarity is critical to the performance of the selection phase, which in turn will determine the performance of a classifier. Results over various datasets show that Fuzzy Laplace Similarity improves consistently the quality of the fuzzy rough feature selection process and performs well compared to other fuzzy similarity relations.