Object registration for remote sensing images using robust kernel pattern vectors

The registration of multitemperal remote sensing images before and after a disaster is a challenging problem. The main reason is that there is a dearth of homologous features that can be used as tie points for registration of variform objects. This paper proposes a new registration method based on robust kernel pattern vector (KPV) for water objects in the remote sensing images, which have complex deformation before and after a disaster. We show how the variform objects can be precisely registered using their robust kernel pattern vectors (RKPVs). The contribution can be divided into three parts. First, a novel shape descriptor, named as kernel pattern vector (KPV), is defined. Second, a robust kernel principal component analysis (RKPCA) method is proposed, which can not only capture the common pattern of the variform objects but can also act as the Criterion for outlier detection. Finally, a registration approach is derived based on the implicit RKPVs. We demonstrate the power of the proposed approach by comparing it with other existing methods using two real cases: one for lake monitoring in the Jiayu region, and the other for damage mapping of earthquake-induced barrier lake at Tangjiashan (2008 Wenchuan Earthquake). The results show that the method is effective in capturing the common structural pattern of the variform objects before and after a disaster.