Robust Illumination Invariant Texture Classification Using Gradient Local Binary Patterns

The local binary pattern (LBP) method has been widely used in texture analysis because it is excellent property of gray scale invariant. Following the basic idea of the LBP method, many robust versions of the LBP are proposed to class textures, especially the rotational textures. Conventional local binary pattern methods for texture classification are only good for common illumination changes, and the performance deteriorates rapidly in the condition of reflection illumination changes. In this paper, a robust illumination invariant operator, the gradient local binary pattern (GLBP), is present. After elaborately constructing comparisons among a local neighbor set, the GLBP operator not only is good for common illumination changes of texture images, but also performs well on the textures under reflection illumination changes. A simple extension makes the proposed method can also be rotation invariant. Experimental results show that the method outperforms conventional approaches in difficult illumination conditions.