Point-based registration of images strongly depends on the extraction of suitable landmarks. Recently, various 2D operators have been proposed for the detection of corner points but most of them are not effective for medical images that need a high accuracy. In this paper we have proposed a new automatic corner detector based on the covariance between the small region of support around a central pixel and its rotated one. The main goal of this paper is medical images so we especially focus on extracting brain MR image’s control points which play an important role in accuracy of registration. This approach has been improved by refined localization through a differential edge intersection approach proposed by Karl Rohr. This method is robust to rotation, transition and scaling and in comparison with other grayscale methods has better results particularly for the brain MR images and also has acceptable robustness to distortion which is a common incident in brain surgeries. In the first part of this paper we describe the algorithm and in the second part we investigate the results of this algorithm on different MR images and its ability to detect corresponding points under elastic deformation and noise. It turns out that this method: 1)detect larger number of corresponding points that the other operators, 2)its performance on the basis of the statistical measures is better, and 3)by choosing a suitable region of support, it can significantly decrease the number of false detection.