Highly precise two-dimensional sub-pixel estimation method on area-based image matching based on similarity model

摘要

Area-based matching is a fundamental image processing to obtain displacement between images. Also, the similarity interpolation method to estimate sub-pixel displacement is commonly used to enhance resolution. Conventionally, similarity interpolation estimation is performed by assuming that horizontal and vertical displacements are independent. Almost no investigations on estimation error over the conventional method have been done, but it is inferred experientially that sub-pixel estimation with image interpolation or gradient-based methods is more precise than the similarity interpolation method. This paper proposes a novel 2D sub-pixel displacement estimation method based on similarity interpolation through modeling 2D self-similarity. The proposed method requires no "a priori" knowledge over 2D similarity, or images at all. It adopts no iteration. Furthermore, the proposed method requires only slightly higher calculation costs than the conventional similarity interpolation method. The proposed method can obtain more precise estimation not only than the conventional similarity interpolation method, but also than the image interpolation method through comparison of sub-pixel estimation accuracy. Moreover, an experiment using actual images demonstrates effectiveness of the proposed method.