A varied local edge pattern descriptor and its application to texture classification

• An edge descriptor, called VLEP, with arbitrary radius and neighbors is proposed.
• Multi-scale and multi-direction (or multi-resolution) VLEPs are derived.
• Multi-scale, multi-resolution and multi-modal fusion ideas are considered.
• Efficient application of VLEP descriptor to texture classification is shown.

Similar images can be classified by the aid of texture cues among which edge is widely considered as one of the most valuable features. In this paper, we firstly proposed a flexible edge descriptor, called varied local edge pattern (VLEP). Then we apply VLEP to similar texture classification. The proposed VLEP descriptor has multi-scale, multi-direction (or multi-resolution) properties. Because VLEP uses histogram spectrum to describe image information, it is very easy to fuse local binary pattern (LBP) and Zernike moments histogram spectrum features due to their excellent properties and supplementary roles. The fused histogram spectrum features representing the images are classified via the nearest neighbor classifier. Experimental results show that the VLEP-based method can be remarkably superior to other state-of-the-art texture classification methods on the large and comprehensive CUReT and Outex texture database.