Remote sensing image retrieval using morphological texture descriptors

This paper presents the results of applying morphological texture descriptors to the problem of content-based retrieval of remote sensing images. Mathematical morphology offers a variety of multi-scale texture descriptors, capable of computing translation, rotation and illumination invariant features. In particular, we focus on the circular covariance histogram and the rotation invariant points approaches, and test them with the UC Merced Land Use dataset. They are compared against other known descriptors such as LBP and Gabor filters, and are shown to provide either comparable or superior performance despite their shorter feature vector length.