Image registration is a fundamental problem in the area of computer vision and has been extensively researched. Different image registration algorithm has its pros and cons. Unlike most of the algorithms, our approach defines a set of feature points uniformly distributed on the image. The feature points are linked up as nodes in a triangular mesh. The image registration in our approach seeks the corresponding points in the other image, while maintaining the triangular mesh. This approach guarantees the mapped result is smooth, and hence it is ideal for smooth objects. Our approach is also relatively good for images that do not have distinctive features.
Ben Yip, Jesse S. Jin