Image Registration Using Dynamic Log-Polar Transformation

Abstract

Log-polar transformation (LPT) is a well-known tool for image registration. LPT makes rotation and scale change in Cartesian coordinate appearing as translation in log-polar domain. Translation can be obtained using phase correlation techniques. However, in the current LPT methods, when images are sampled, the fixed number of samples tends to result in low precision. Aiming to this issue, this paper presents a dynamic LPT (DLPT) method, which allows the number of samples to change dynamically within a scope and generates a series of results. Among the set of results, we remove the invalid ones and those with large bias values, i.e., only those with small bias values can be retained, and then we make average of the remained results to get the final result. Experimental results show that DLPT is more powerful when compared with some available methods.