Abstract

Retrieval of remote sensing image from the large database is the difficult task. To solve this problem in this paper we are presenting the multiscale descriptors like circular covariance histogram, rotation invariant point triplets are used which are the effective texture descriptors. To overcome some of the limitations of the circular covariance histogram and rotation invariant point triplet, we introduce new descriptors Fourier power spectrum and quasi flat zone which consider the important texture properties like coarseness and directionality in the image which is the important property of the remote sensing images. The proposed system achieves best retrieval performance and also reduces the length of the feature vector. UC Merced land use land cover data set is used to evaluate the effectiveness of the descriptors which is the recently introduced dataset.