Effect Analysis in the Fine Co-Registration of Very-High-Resolution Satellite Images for Unsupervised Change Detection

Fine co-registration that precisely aligns multiple images acquired over a given area is an important process to exploit the very high resolution (VHR) multitemporal images in a wide range of remote sensing applications. The objective of this study is to analyze the effect of the fine co-registration performance on an unsupervised change detection between VHR images. To this end, we extract registration noise (RN) samples, which are denoted as misaligned pixels in a local region. Then, the location of conjugate points (CPs) is positioned by analyzing the local distribution of the extracted RN samples. The CPs are employed for generating a non-rigid transformation model to warp a sensed image into a reference image. An unsupervised change vector analysis approach is used to validate the effectiveness of the proposed fine co-registration performance. Experiments are implemented on a Worldview-3 VHR multispectral dataset.