Ear biometrics is a fast-developing field of computer vision that offers several advantages over existing biometric authentication methods. The performance of the ear is superior to or comparable to that of other biometric traits. However, the recognition performance of 2D ear images degrades due to the impacts of posture, illumination, and scaling. To overcome these challenges, we propose combining 3D and 2D ear images and constructing a keypoint detector and descriptor using a covariance matrix derived from texture and geometry information. The feature vector of the descriptor is then used to describe keypoints and utilized to identify keypoints with the smallest distance between a probe and gallery for the subsequent registration process. The registration error that results serves as the matching score. The proposed method is evaluated on a large database of ear images and compared to state-of-the-art techniques, demonstrating superior performance in occlusion, noise, and pose variations.