Journal of Pattern Recognition Research

The local binary pattern (LBP) is a gray-scale and rotation invariant operator, and has been proved to be theoretically very simple, yet computational efficient approach for texture classification. As for the irregular texture surface image, like pavement surface image, the original LBP performs not good enough for practical purposes. First, threshold in plain area often cause mismatch in local structure. Second, nonuniform patterns were directly merged into one pattern will discards large amount of texture information represented by these patterns. In this paper, a novel LBP based operator for pavement crack detection is proposed. In our approach, local neighbors are classified into smooth area and rough area, segmentation only performed in rough area to catch local structure information. And then, local patterns are regrouped and a lookup table is created for fast implement. With these methods, the proposed approach detects cracks well and becomes more robust against noise. Experiments on the pavement surface image show the good performance of this new LBP based operator. More importantly, because of its simplicity, online implementation is possible as well.