6.1.1 Synthetic dataset

The synthetic and medical dataset containing 720 images are artificially created from images reported in [10, 21]. There are actually about 20 images reported in [10, 21]. To avoid over-fitting problem as well as make sure the proposed RLS is able to deal with both intensity homogeneity and inhomogeneity, different kinds of degradation and affined have been used. As for degradation, we use many kinds of noise and blurring at different ratio. As for affine transformations, we use rotation, translation, scale, and flip at different degrees. As a result, we obtain 720 images on this dataset. Half of this data set is used for training, i.e., 360 images which were artificially generated from first 10 images. The rest of 360 images is used for testing. Furthermore, we use the one object dataset from Weizmann [22] for the conducting the experiment on natural images. Using the same strategy of augmentation as in medical images, we obtain 4700 images and we used 2350 images for training and 2350 images for testing.
