A Context-Sensitive Technique Robust to Registration Noise for Change Detection in Very High Resolution Multispectral Images

Abstract

In this paper an automatic context-sensitive technique robust to registration noise (RN) for change detection on multitemporal very high geometrical resolution (VHR) images is presented. Exploiting the properties of RN in VHR images, the proposed technique analyzes the distribution of the spectral change vectors (SCVs) computed according to the change vector analysis (CVA) in a quantized polar domain. The method studies the SCVs falling into each quantization cell at different resolution levels (scales) to automatically identify the effects of RN in the polar domain. In order to improve the change-detection accuracy also the spatial-contextual information contained in the neighborhood of each pixel is considered through the definition of adaptive regions homogeneous both in spatial and temporal domain (parcels). The final change-detection map is generated considering both the information from the multiscale analysis and the spatial-contextual information. Experimental results obtained on real VHR multitemporal images confirm the effectiveness of the proposed approach.