Fan-shaped patch local binary patterns for texture classification

In this paper, we present a new distinctive feature for texture classification, the fan-shaped patch local binary patterns (FP-LBP). The proposed FP-LBP operator extends the traditional LBP operator by encoding the difference between each central pixel with the average value of its neighboring fan-shaped patches, instead of only using its neighboring pixels. By this way, FP-LBP not only preserves more information of local structures than the traditional LBP, but also keeps relatively lower dimensionality, especially when larger radius and more neighboring pixels are considered. Moreover, the “uniform” and rotation invariant FP-LBP are also defined similarly to the traditional LBP. The proposed descriptors are evaluated on two popular texture databases: CUReT and KTH-TIPS, and the experimental results show that FP-LBP outperforms the traditional LBP descriptor with a smaller feature dimension. Moreover, the proposed method achieves higher classification accuracy than most of the state-of-the-art methods on both databases.