A probabilistic framework for dense image registration using relaxation labelling

Image Registration has investigated in many researches in recent years. It is an important preprocessing step in a variety of applications such as medical images, super resolution and remote sensing. Generally, dense image registration requires several transformations and deformations such as contrast changing, scaling, rotation and displacement. However, in most recent proposed methods, only some of these transforms are considered, which results in incorrect output. In this paper we propose a new method for dense image registration based on relaxation labelling. For each pixel of a test image, we want to find the best match in a reference image, considering the intensity and geometric constraints. We use blocks of reference image as features, then we look for the closest candidate in the test image. In next step, a relaxation labelling procedure is applied to these candidates for selecting the best match between candidate pixels. Experimental results show that the proposed method achieves satisfactory performance in terms of visual quality, PSNR values and Bad pixel evaluation criteria.