What is Image Registration?
Image registration geometrically transforms two or more images of the same subject into a common coordinate system. The inputs may differ by sensor, capture time, depth, viewpoint, or imaging method.
How Image Registration works
Registration estimates a spatial transform that maps a moving image onto a chosen reference. The estimate may come from matched landmarks, intensity similarity, sensor geometry, or a combination, and its allowed motion can range from rigid to locally deformable. The transformed image is then resampled on the reference grid. This alignment step precedes mosaicking, multisensor fusion, temporal comparison, medical measurement, and other workflows that combine observations.
Key facts
- Rigid transforms preserve distances and angles, affine transforms also permit scale and shear, and projective transforms model perspective on planar scenes.
- Multimodal images may not share comparable brightness values, so registration can optimize statistical dependence or structural features instead of raw pixel differences.
- Repeatedly resampling intermediate results compounds blur and aliasing; composing transforms and sampling the original once better preserves measured image content.
When Image Registration matters
Choose rigid registration when only translation and rotation are expected, and a deformable method when the subject itself can change shape. Incorrect feature matches can produce plausible-looking alignment that invalidates later comparison.