240 likes | 267 Views
Learn about hierarchical B-Splines for image fusion, comparison, segmentation, and pattern recognition in image registration. Explore point-based, curve-based, and intensity-based methods, along with Free-Form Deformation (FFD) techniques.
E N D
Image Registration with Hierarchical B-Splines Z. Xie and G. Farin
Support Arizona Alzheimer Disease Research Center
Motivation • Image Fusion • Image Comparison • Image Segmentation • Pattern Recognition
Classification • Landmark based methods • Point based method • Curve based method • Surface based method • Intensity based methods
FFD with Hierarchical B-Splines • Putting the object into the B-Spline hyperpatchs. • Moving the B-Spline control points to deform the object. • Refining the control points related to complex regions. • Adjusting the refined control points for detail deformation.
Point based registration This problem naturally breaks down into two scattered data approximation problems. The least squares solution of this problem can be found by solving the linear systems.
How does it work? • Local refinement by knot insertion. • Recomputing related control points.
Why hierarchical B-Splines? • Efficiency • Global to local influence
Example of point based registration Source Target Deformed Source
Surface based registration Together with the Iterative Closest Point (ICP) approach, this problem can be converted into a scattered data approximation problem.
Hierarchical Deformation with Hierarchical B-Splines • Initialize: Rigid Transformation • Linear matching: Iterative Affine Deformation • Nonlinear matching: Hierarchical Cubic B-Splines • Increase level of detail iteratively
Advantage • Validity. Right matching between individual points by matching big shape feature first, then refine the detail gradually. • Efficiency. Only pay attention to complex regions. • Precision. Enough of degrees of freedom for matching detail.
Intensity-based registration Together with optic flow, this problem can be converted into scattered data approximation problem.
Optic flow Optic flow is a visual displacement flow field associated with the variation in an image sequence. It can be used as an estimator of the displacement of one pixel on the source image to its matching pixel on target image.
Hierarchical Deformation vs. multi-resolution data representation
Example of intensity based registration Source Deformed source Target
Future Work • Multi-resolution surface representation • More robust displacement estimator for intensity based registration. • Multi-modal intensity based registration