Scale-invariant Texture Description using Local Patch Vectors

In this paper, a new scale-invariant texture descriptor is presented. In the proposed method, the local patch vectors are implemented in a scale-space structure to extract scale-invariant features from texture images. The main idea of the proposed method is that the Euclidean norm of the local patch vectors in large scale textures are smaller than that of fine textures. Therefore, an appropriate vector can be selected in a scale-space structure for each pixel. Considering the robustness of local patch distance vector (LPDV) and local patch angle vector (LPAV) to noise, rotation and illumination changes, the proposed method in this paper results in a more effective texture descriptor. The experiments in classification of Outex and Brodatz datasets with different non-ideal conditions such as noise, rotation, illumination and scale changes showed superior performance of the proposed method in comparison with conventional texture descriptors.

Keywords:

texture description , feature extraction , scale-invariant , local features.