Texture Characterization in Remote Sensing Imagery Using Binary Coding Techniques

In this paper rotation invariant Local Binary Patterns (LBP) texture based descriptors are evaluated experimentally in the context of land-use and land-cover object-based classification. The texture descriptors were employed in the classification of an Ikonos-2 and a Quickbird-2 image. The experiments have shown that texture characterization approaches perform well when combined with the grayscale variance. We further investigate a novel descriptor resulting from the concatenation of the grayscale variance histogram and the histogram of codes generated by LBP. These experiments have demonstrated that the proposed descriptor, though more compact, performs as well as a bidimensional histogram representing the joint distribution of both quantities. Finally, the paper compares the discrimination capacity of the LBP based textural descriptors with that of features derived from the Gray Level Cooccurrence Matrices (GLCM). The related experiments revealed a noteworthy superiority of LBP descriptors over the GLCM features in the context of remote sensing image data classification.