A contextual change detection method for high-resolution optical images of urban areas

Several Earth observation satellites acquire images with different spatial resolutions, e.g., a panchromatic band with high resolution, and spectral bands with lower resolution. Likewise, we often face the problem of the presence of different resolutions when performing joint analyses of images acquired by different satellites. This paper presents a methods for unsupervised change detection of multiresolution images. The approach is based on the concept of a reference resolution, corresponding to the highest resolution in the dataset. The spatial relationships between the class labels are specified through a Markov random field model defined at the reference resolution. Data at coarser scales are modelled as mixed pixels by relating the observations to the classes at the reference resolution. A Bayesian framework for classification based on this multiscale model is adopted. The classification is performed by an "iterative conditional modes" (ICM) algorithm. A computationally efficient scheme for parameter estimation based on a combination of the ICM and of the expectation-maximization algorithm is used. Results obtained on real satellite images are presented.