Registration of challenging image pairs: initialization, estimation, and decision.

Area of Science:

• Computer Vision
• Image Processing
• Geometric Transformations

Background:

• Automated image registration is crucial for various applications, including medical imaging and scene analysis.
• Existing methods struggle with significant variations in overlap, scale, orientation, and illumination.

Purpose of the Study:

• To develop an automated 2D image-pair registration algorithm.
• To handle challenging conditions like low overlap, scale/orientation differences, and illumination variations.
• To automatically reject non-registrable image pairs.

Main Methods:

• Keypoint matching for initial similarity transform estimation.
• Refinement using the Dual-Bootstrap Iterative Closest Point (ICP) algorithm with multiscale features.
• A three-part decision criteria for validating transformation estimates (accuracy, stability, consistency).

Main Results:

• Successfully aligned 19 out of 22 challenging image pairs.
• Achieved 99.8% rejection rate for misalignments across all possible pairs.
• Demonstrated superior performance compared to keypoint matching alone.

Conclusions:

• The proposed algorithm offers robust and automated 2D image registration.
• It effectively handles significant image variations and reliably rejects poor matches.
• This advancement has broad applicability in fields requiring accurate image alignment.