Geometrical image features like edges and ridges in digital images may be extracted by convolving the images with appropriate derivatives of Gaussians. The choice of the convolution operator and of the parameters of the Gaussian involved defines a specific feature image In this paper, various feature images derived from CT and MR brain images are defined and tested for usability and robustness in a correlation-based two and three dimensional matching algorithm. A number of these feature images is shown to furnish accurate matching results. The best results are obtained using gradient magnitude edgeness images.