Registration Using Robust Kernel Principal Component for Object-Based Change Detection

Abstract

Postclassification comparison provides a feasible approach to detect the changes of remote sensing images which have strongly inhomogeneous scenes. For pre- and postevent scenarios, registration is a challenging task because variform classifications may result in a dearth of homologous points to be used as tie points. In this letter, we show how the variform objects can be precisely registered using their robust kernel principal components (RKPCs). The contribution can be divided into two parts. First, a robust kernel principal component analysis (RKPCA) method is proposed to capture the common pattern of the variform objects. Second, a registration approach based on the implicit RKPCs is derived. We demonstrate the power of the proposed approach 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. The results show that the method is effective in capturing structural pattern and generalizes well for registration.