Rotation, scale and translation invariant automatic target recognition based on Fourier-Mellin transform and bispectrum for satellite imagery

Abstract

Automatic target recognition for satellite imagery is approached in this study by using template matching. Three methods are exploited for rotation, scale and translation invariance of template matching and compared in performance. These methods utilize Fourier-Mellin transform directly, a reconstruction-based method via Fourier transform, log-polar mapping and phase correlation and finally, a reconstruction-based method by using bispectrum, log-polar log-polar mapping and phase correlation. Acceptable performance rates are obtained for the reconstruction-based methods. It is also observed that the method using bispectrum has the potential for highest performance rates, while it can suffer from the drawback of slice constant selection.