Feature discriptor matching Flashcards
(23 cards)
What is a feature descriptor?
A compact numerical representation of a local image region used for matching keypoints across images.
What properties make a good descriptor?
Distinctiveness, invariance (scale, rotation, lighting), compactness, and repeatability.
What does SIFT stand for?
Scale-Invariant Feature Transform.
What are the key steps in SIFT descriptor construction?
Detect scale-space extrema, assign dominant orientation, divide patch into 4×4 cells, compute 8-bin orientation histograms per cell.
How many dimensions does a SIFT descriptor have?
128 dimensions (4×4 cells × 8 orientation bins).
What kind of invariance does SIFT offer?
Scale, rotation, and partial illumination invariance.
What distance metric is used to compare SIFT descriptors?
Euclidean distance (L2 norm).
What is the ratio test in feature matching?
A rule that accepts a match only if the best match is significantly better than the second-best match (e.g., distance₁ / distance₂ < 0.8).
What problem does the ratio test address?
Ambiguous matches caused by repetitive textures or weak features.
What is the correspondence problem in image matching?
The challenge of identifying correct keypoint matches in the presence of noise, occlusion, or similar features.
What is RANSAC used for in feature matching?
To robustly estimate a transformation between matched points by rejecting outliers.
What does RANSAC stand for?
Random Sample Consensus.
What is the main idea behind RANSAC?
Repeatedly sample minimal subsets to fit a model, then select the model with the largest inlier set.
What is an inlier in RANSAC?
A match that fits the estimated model within a small error threshold.
What model types can be estimated using RANSAC?
Affine transformation, homography, fundamental matrix, etc.
How many points are needed to estimate an affine transformation?
At least 3 point correspondences.
What is the general form of a 2D affine transformation?
[x’, y’]ᵀ = M × [x, y]ᵀ + t, where M is a 2×2 matrix and t is a translation vector.
What are the benefits of using feature descriptors over raw pixel matching?
Greater robustness to scale, rotation, lighting changes, and partial occlusion.
What is the difference between descriptor matching and direct pixel matching?
Descriptor matching compares compact, invariant vectors at keypoints; direct matching compares raw pixel intensities.
What are the typical components of a descriptor matching pipeline?
Keypoint detection, descriptor construction, matching, match filtering, and robust model fitting.
What is FLANN used for?
Fast approximate nearest neighbor search for matching high-dimensional descriptors like SIFT.
Why are binary descriptors faster to match?
Because Hamming distance can be computed quickly using bitwise operations.
What is descriptor normalization and why is it done?
Scaling a descriptor to unit length to reduce sensitivity to lighting and contrast variations.
