A Brief Study on a Novel Texture Spectrum Descriptor for Material Images

Abstract

The local binary pattern (LBP) operator has been proved to be theoretically simple yet very effective for texture description. However, it lacks the full description of texture in a region and produces a rather long histogram. A novel texture spectrum descriptor was proposed to alleviate these limitations in the paper. It uses the relation of 3 pixels in an 8-neighborhood, the center and the center-symmetric pixels, to define the local texture patterns. The new operator fully uses the texture information contained in the 8-neighbourhood and produces a rather short histogram. On the other hand, the new operator also has the same desirable properties as LBP, such as tolerance to illumination changes and computational simplicity. Experimental results demonstrate that the new descriptor achieves better performance than the conventional LBP with a rather short histogram