The representation, recognition, and locating of 3-d objects
International Journal of Robotics Research
A Method for Registration of 3-D Shapes
IEEE Transactions on Pattern Analysis and Machine Intelligence - Special issue on interpretation of 3-D scenes—part II
Object recognition based on moment (or algebraic) invariants
Geometric invariance in computer vision
Pattern Recognition
| Hi-index | 0.00 |
A new weighted quaternion based non-rigid registration is presented in this paper. Strong crest points derived from principal curvatures provide the most robust features for image registration. Crest point strengths are based on their principal curvatures and the number of scales a particular crest point is detected at. Geometric features are extracted which are invariant to rotation, translation and scaling by using neighborhood crest points only as other voxels are susceptible to deformation. The neighborhood size is adjusted according to scale adaptively using a fixed k nearest neighbor to make the extracted feature scale invariant. Statistical properties are used to measure the distribution of these geometric invariant features. The scale and rotation invariant feature points are then used to establish a point to point correspondence between the template crest points and the subject image crest points. A multi-scale feature based subdivision scheme is employed for registration where a weighted quaternion matrix provides a quaternion transformation based on the corresponding points to obtain the best rotation for global as well as local sub-blocks.