For comparison, We also include in Table . the recog-nition performance of the 3D-SIFT and 3D-HOG features obtained from the 'bag-of-features'(BoF) + linear SVM, BoF+3-nearest neigbors (3NN) and Random Forest classifiers. Here, and elsewhere in this paper, We have constructed a BoF codebook using K-means clustering with 200 centures with an NBNN classifier significantly outperforms the other schemes.