Very High Resolution (VHR) multitemporal images that have been already applied registration process still show a residual misalignment due to the dissimilarities of the acquisition environment. The objective of this paper is to mitigate the residual misalignment between VHR images to get a fine geometric alignment result. Here we propose to refine the local misalignment between VHR images by extracting Registration Noise (RN), which is denoted as misaligned samples. Extracted RN pixels are used as Control Points (CPs), and local distribution analysis of the CPs in a specific region defined by a quadtree structure is carried out for finding their correspondences. Matched CP pairs are employed for generating a deformation map to warp the sensed image to the master image. Experiments carried out on both simulated and real multitemporal VHR datasets acquired from IKONOS and QuickBird sensors confirm the validity of the analysis of the proposed method.