Fig. 5From: Exploiting textures for better action recognition in low-quality videosPercentage improvement of BSIF-TOP over LBP-TOP and LPQ-TOP when combined with STIP on downsampled KTHBack to article page