Improving Rotation Invariance of the Volume Local Binary Pattern Operator

Dynamic texture is an extension of texture to the temporal domain. Recently, a powerful method for dynamic texture recognition based on volume local binary patterns (VLBP) was proposed. In this paper, we investigate improvements of the original VLBP operator. A proof on the relation of the two rotation invariant VLBP patterns is given. Methods for obtaining rotation invariance are experimentally evaluated with differently rotated dynamic texture sequences. An approach for combining uniform and non-uniform local binary patterns is also proposed. Experimental results on DynTex database show the effectiveness of the approach, especially its robustness to rotation variations and effective compression of feature vectors.

• Type: Conference
• Year: 2007
• Where: MVA
• Authors: Guoying Zhao, Matti Pietikäinen
• Keywords: Computer Vision, Dynamic Texture, Dynamic Texture Recognition, Local Binary Patterns, MVA 2007