A novel segmentation-based mutual information image registration for SAR images

In this paper, a segmentation-based mutual information image registration algorithm is proposed. Because of noise-like speckle, mutual information is calculated on segmented results instead of original images. Instead of using pixel-based MRF segmentation, region-based RAG-MRF is applied on oversegmentation formed by Watershed transformation on the gradient of SAR images. The extrema of mutual information indicate the desired registration parameters. The conception of mutual information image registration is introduced followed by the principle of RAG-MRF segmentation and watershed transformation. Experiment results showed it is an efficient approach for registration of SAR image.