Discriminative Dictionary Design for Action Classification in Still Images and Videos

In this paper, we address the problem of action recognition from still images and videos. Traditional local features such as SIFT STIP invariably pose two potential problems: 1) they are not evenly distributed in different entities a given category 2) many exclusive visual concept represent. order to generate dictionary taking aforementioned issues into account, propose novel discriminative method for identifying robust specific which maximize class separability greater extent. Specifically, selection potent descriptors filtering-based feature problem, ranks per based on measure distinctiveness. The underlying subsequently represented learned dictionary, stage is followed by classification using random forest model label propagation refinement. framework validated datasets (Stanford-40) well videos (UCF-50). We get 51.2% 66.7% accuracy Standford-40 UCF-50, respectively. Compared other representative methods literature, our approach exhibits superior performances. This proves effectiveness adaptive ranking methodology presented work.