A Robust and Invariant Keypoint Extraction Algorithm in Brain MR Images

In this paper a method for extracting keypoints from human brain MR images is proposed. These keypoints are obtained based on curved structures in the brain MR images. In this method, a keypoint is center of a circle which includes circular boundaries in the image and is selected based on gradients of the image. These keypoints and their descriptors are scale and rotation invariant. The proposed method is compared with other well-known methods with repeatability measure and ROC curves. Experimental results show that proposed method performs better than other well-known methods, specially, when deformations are remarkable.