Non-homogenous Auto-regressive Model Based Prior for Multiresolution Fusion

1. ABSTRACT In this paper we propose a model-based approach for multiresolution fusion of remotely sensed images. Given a high spatial resolution & low spectral resolution PAN (Panchromatic) image and a low spatial resolution & high spectral resolution MS (Multispectral) images of the same geographical area, the objective is to enhance the spatial resolution of the MS images to that of the PAN image i.e. to obtain a high spatial and spectral resolution images. A proper regularization technique is required to address this ill posed problem and get a better solution. We use a nonhomogenous AR (auto-regressive) model based prior for each of the fused MS images. This method is insensitive to registration errors between PAN and MS images unlike other methods. The AR parameters are estimated using the segmented regions of the PAN image. The estimated AR parameters are then used in minimizing the cost function. Experimental results are illustrated for Landsat-7 data set.