image matching Flashcards
(29 cards)
What is image matching?
The process of identifying corresponding points or regions between two or more images.
What are applications of image matching?
Mapping, 3D reconstruction, medical imaging, SLAM, quality control, and damage assessment.
What is the goal of stereo matching?
To find corresponding points in stereo images that represent the same 3D scene location.
What is the stereo normal case?
A setup where cameras are aligned horizontally, simplifying disparity calculation.
What are two approaches to finding corresponding points?
Manual matching and automatic matching.
What does automatic matching require?
A suitable descriptor and a matching criterion.
What is template matching?
A method to locate a small image patch (template) within a larger image using similarity scores.
What is cross correlation (CC)?
A method for measuring similarity between an image and a shifted template using pixel intensities.
What is the formula for cross correlation?
CC(u) = ∑ I₀(x) · I₁(x + u)
What is normalized cross correlation?
A version of CC that accounts for local brightness and contrast variations.
What is the output range of normalized cross correlation?
Between -1 and 1, where 1 indicates perfect match.
What is an exhaustive search in CC?
Testing every possible offset to find the best match, which is computationally expensive.
What is a coarse-to-fine search strategy?
Using image pyramids to first match at low resolution and refine at higher levels.
What are the limitations of cross correlation?
Fails under large rotations, scale changes, occlusion, and lighting variation.
What is least squares matching (LSM)?
A model-fitting approach that estimates transformation parameters by minimizing intensity errors.
What is the objective of least squares fitting?
To minimize the sum of squared differences between predicted and observed values.
What type of transformation can LSM estimate?
Affine transformations such as rotation, scaling, translation, and shear.
How many point correspondences are needed to solve an affine transformation?
At least 3, giving 6 equations to solve for 6 unknowns.
What is the general form of a 2D affine transformation?
[x’, y’]ᵀ = M · [x, y]ᵀ + t
What are the strengths of least squares matching?
High accuracy and flexibility for parametric models.
What is feature-based image matching?
Using keypoints and descriptors to find correspondences between images.
What are common descriptors in feature-based matching?
SIFT, SURF, ORB
What are the advantages of feature-based matching?
Robust to scale, rotation, and partial occlusion.
What is direct matching?
Pixel-by-pixel alignment using intensity differences.
