Robust registration of satellite images with local distortions

A new method for precise registration of satellite images in the presence of local distortions is presented. The template matching is based on Cumulative Residual Entropy (CRE) of the difference-image, a new and powerful similarity measure. The regularization process relies on inpainting, technique that has demonstrated eccellent results in frame interpolation or in improving sparsely sampled data. The CRE measure is simple to estimate, robust to noise and local deformations, invariant to amplitude scaling, more efficient on small sample data than other similarity measures such as mutual information and cross-correlation. An outliers detection method is applied to locate potential errors, that are rejected and corrected with the inpainting algorithm. Experimental results on simulated images have shown quantitatively the accuracy of the proposed method.