Change Detection in VHR Multispectral Images: Estimation and Reduction of Registration Noise Effects

Abstract

In this chapter we address the problem of change detection (CD) in very high geometrical resolution (VHR) optical images by studying the effects of residual misregistration (registration noise) between images acquired over the same geographical area at different times. According to an experimental analysis driven from a theoretical study, we identify the main effects of RN in VHR images and derive some important properties exploiting a polar framework for change vector analysis (CVA). On the basis of the identified properties, we propose: (i) a technique for an adaptive and unsupervised explicit estimation of the RN distribution based on a multiscale analysis of the behavior of spectral change vectors in the polar domain and the Parzen window method; and (ii) an automatic context-sensitive technique robust to registration noise (RN) for CD based on a multiscale analysis in a quantized polar domain. Experimental results obtained on simulated and real VHR multitemporal images confirm the validity of the proposed analysis on RN, the reliability of the derived properties and the effectiveness of the proposed techniques for the estimation of RN distribution and change detection.