Texture description in local scale using texton histograms with quadrature filter universal dictionaries

An inherent property to the texture patterns is that they are only meaningful in an appropriate range of scales. Taking this into account, the description of the texture patterns should be limited to its meaningful scales. This assumption motivates the research on local-scale texture description. In this study, a method for the extraction and description of texture features using local scale is presented. The method is based on texton histograms using universal prototype dictionaries of extremal quadrature filter outputs. Comparison of results with state-of-the-art texture description methods demonstrate the higher discriminative power of the proposed method in local-scale texture classification.