Very high resolution image registration based on two step Harris-Laplace detector and SIFT descriptor

Abstract

At present, registration of the Large size Very High Resolution (VHR) images is one of the important task in remote image analysis. However, until now, it is still rare to find an accurate and robust image registration method, and most of the existing methods are designed for small size images. Among the most popular methods, SIFT is performed well to register VHR images but the calculation procedure is too slow. To overcome this difficulty coarse-to-fine strategy and parallel implementation have been proposed but again this increased the extra overhead of multiple PCs. This paper proposes a coarse-to-fine strategy, that combine Harris-Laplace detector together with SIFT descriptor. Proposed approach starts with roughly register the image at first place and then to perform fine registration by dividing the large size image in to small size process-able blocks. It improves registration accuracy by providing increased matching ratio as well as makes computation fast without the need of multiple PCs.