A Fourier-Based Global Descriptor For Image Matching

Fourier Coefficients have long been used to achieve invariance to signal transformations. In this paper we propose a Global Rotation Invariant Descriptor (GRID) for matching images based on the Discrete Fourier Transform (DFT). Following the analysis of the sampling criteria for generating such descriptors, it is shown that the descriptor can be made invariant to changes in scale as well under certain conditions of preserved aspect ratio and image boundaries. Furthermore, it is shown that the effects of common compression algorithms is negligible on the performance of the descriptor. The performance and efficiency of the proposed global descriptor is then compared to that of an alternative local solution. It is shown that the proposed global solution significantly outperforms the local one in terms of speed and efficiency with acceptable compromises in the accuracy of the matching performance.