COMPARATIVE ANALYSIS OF SUBPIXEL ALGORITHMS FOR IMAGE MATCHING

V.A. Dushepa, M.L. Uss

In this study, a comparative analysis of subpixel algorithms for image matching using normalized cross-correlation in image-based navigation systems is conducted. Three subpixel algorithms (intensity interpolation of the reference image, correlation coefficient curve-fitting (interpolation) and gradient-based method) are considered. The algorithms were investigated on artificially simulated textured images with different smoothness and a real image of hyperspectral radiometer AVIRIS at different spatial displacements between a current image and a reference image. According to simulation results, the intensity interpolation method has the best accuracy. The accuracy analysis of considered algorithms in comparison with the Cramer-Rao lower bound is conducted. It is shown that this bound becomes inadequate for describing real accuracy of considered algorithms when the signal-to-noise ratio is large.

Key words: image matching, subpixel accuracy, normalized cross-correlation, image-based navigation, Cramer-Rao bound.