Keypoint detection is a well-established technique in computer vision that finds applications in various domains, particularly in tasks such as image registration and scene stitching, where high accuracy is crucial. One prominent challenge in these applications is the accurate handling of object rotation. The standard image registration process involves keypoint detection, description, and matching. Recent advancements in hardware capabilities have spurred research efforts to enhance keypoint detection by applying deep learning methodologies. However, relying solely on deep learning falls short of addressing rotations in three-dimensional space. To tackle this, we present a novel algorithm focused on keypoint matching. The process involves keypoint detection, descriptor computation, and the formulation of multi-angle descriptors using a three-dimensional projection technique. In empirical evaluations, our proposed method demonstrates significant improvements in accuracy, ranging between 5% and 20%, across various keypoint detection models and datasets. Impressively, these enhancements come without substantial computational overhead, making our approach compatible with diverse keypoint detection models and hardware configurations.