The need for accurate and efficient computation of template matching prevails in many applications. However, a challenge in template matching is to obtain high accuracy that involves acceptable computation complexity and is robust to rotation. A new subpixel template matching approach that combines the parametric template method and the ring projection transform process is proposed. It not only achieves subpixel accuracy in location, but also offers rotation invariance in the subpixel template matching. Furthermore, our approach is conceptually simple, easy to implement, and very efficient because no iterative steps are involved. The simulated results show that our approach enjoys very high precision in the presence of image rotations. Experiments with real-world scenes demonstrate that the proposed method can reach subpixel accuracy for finding the distance between two target objects in the presence of rotations and translations. This indicates that our approach is suitable for accurate on-line template matching with scene rotations and translations.
All Science Journal Classification (ASJC) codes
- Atomic and Molecular Physics, and Optics