Return of the king: The Fourier transform based descriptor for visual object classification

Most of the state-of-arts visual object classification methods use bag of words model for image representation. In this method, patches extracted from images are described by different shape and texture descriptors such as SIFT, LBP, SURF, etc. In this paper we introduce a new descriptor based on weighted histograms of phase angles of local Fourier transform (FT). We compare the classification accuracies obtained by using the proposed descriptor to the ones obtained by other well-known descriptors on Caltech-4 and Coil-100 data sets. Experimental results show that our proposed descriptor provides good accuracies indicating that FT based local descriptor captures important characteristics of images that are useful for classification. When we combined image representations obtained by FT descriptor with the representations obtained by other descriptors, results even get better suggesting that tested descriptors encode differential complementary information.