Dynamic texture recognition based on multiple statistical features with LBP/WLD

Dynamic texture is an extension of traditional texture to spatial and temporal domain. Description and recognition of dynamic texture has attracted much attention recently. Local Binary Motion Pattern (LBMP) based on Local Binary Pattern (LBP) is easier to compute and outperforms other methods. However, the matching rule employed in LBMP is not accurate enough as well as the extracted feature is not comprehensive. This paper presents a dynamic texture recognition method based on multiple statistical characteristics with LBP and Weber Local Descriptor (WLD). Experimental results on Dyntex and Szummer database show that this method has better recognition accuracy than LBMP.