Image Registration

Interactive Demonstrator · Medical Image Analysis · Uni Bern
Registration Explorer
Apply transforms to the moving image and watch similarity metrics respond in real time. Try aligning the images to maximize NCC / MI, or minimize SSD.
Fixed I
Moving T(J)
Checkerboard
|I - T(J)|
SSD (lower=better) NCC (higher=better) MI (higher=better) | dashed = current
Modality
Transform
Translate X 0
Translate Y 0
Rotation 0
Metrics
SSD ↓0
NCC ↑0
MI ↑0
Matrix
Mono-modal: identical contrast. SSD/NCC/MI all find the optimum.

Multi-modal: inverted contrast (like CT vs MRI). Only MI finds the correct alignment — SSD has no useful minimum.

Try Misalign, then use sliders to re-align. Which metric guides you best?
Iterative Closest Point (ICP)
Step through ICP: find correspondences, solve Procrustes, update. Watch the cost decrease.
Target Y Source X Correspondences
Controls
Initialization
Iteration0
Mean Error--
Convergence
ICP alternates nearest-neighbor matching with Procrustes.

Cost decreases monotonically but may hit a local minimum with poor init. Try it!
Interpolation Viewer
Rotate a pattern and compare Nearest Neighbor vs Bilinear side by side. Zoom in to see artifacts.
Nearest Neighbor
Bilinear
Controls
Rotation 15
Zoom 3x
Pattern
Nearest neighbor: picks closest pixel. Blocky. Good for label maps.

Bilinear: blends 4 neighbors. Smooth. Default for intensity images.

Increase rotation and zoom to see artifacts clearly.