A generalized local binary pattern descriptor and its applications to texture retrieval

In this paper, we develop a rotation-invariant (RI) texture descriptor, termed generalized local binary pattern (GLBP). The proposed GLBP captures more detailed information by integrating local features in inter-directions and relative distribution among intra-directions. GLBP is characterized by extracting LBP-sign and LBP-magnitude to describe local structures jointly, and performing Fourier transform on both of them to obtain global RI features. Furthermore, a GLBP-based texture retrieval approach is applied to well-known Brodatz texture database and its rotation database. Experimental results demonstrate that GLBP is an effective texture descriptor.