Pattern Matching Using Partial Fourier Spectrum

Pattern matching is a fundamental technique in image processing. The majority of improvements that have been proposed in the literature are geared toward small- to mid-size templates and images. The rapidly increasing resolution and size of images demand a more efficient template-matching algorithm. The main focus of this paper is to address this problem by proposing a novel template-matching algorithm using partial Fourier spectrum (PFS) components as its feature set. It is shown that regardless of the content of the template in terms of correlation in spatial domain, the use of PFS components results in a smooth error surface, which facilitates the evaluation of the region of interest by predicting the error change within the foreseeable distance. It is also shown that this approach generates a broad global minimum area around the best match, which is on the order of the size of the template. Based on these characteristics, three algorithms are developed and the performance of the algorithms is analyzed through extensive mathematical analysis and experimental simulations. The experimental results, generated using over 13 000 image/template pairs, indicate that the proposed algorithms outperform the most efficient algorithms currently available in the literature by over an order of magnitude.