A Novel Spatial-Scale Weighted GIST Descriptor for SAR Image Retrieval

Abstract

In this paper, a novel spatial and scale weighted GIST (SSWGIST) descriptor is proposed for synthetic aperture radar (SAR) image retrieval. Motivated by GIST features, images are represented by the mean values of adjacent and non-overlapped blocks of the Gabor filters response. Beyond that, our methods give those values different weights on different spatial and scale sites. The spatial weights are obtained adaptively by counting the ratio and significance of edges detected in the blocks. The scale weights obey the Gaussian distribution with special parameters toward given image datasets. Thus, the prominent identity of each block of filtering response can be reflected adaptively. A retrieval scheme experiment is carried on the Brodatz and SAR image datasets. The results reveal our algorithmpsilas efficient performances and superiorities.