Descriptors and Matching
SIFT, ORB, and HOG descriptors, the distances that compare them, and the tests that filter matches. Quiz 1 this session.
Key Terms
SIFT DescriptorORB / BRIEFHOGL2 & Hamming DistanceRatio TestCross-Check
Python Tools
OpenCVNumPyscikit-imageMatplotlib
Textbook Reference
Szeliski Ch.7
Topics Covered
- 1SIFT descriptor: orientation assignment, gradient histograms, normalization, and clipping
- 2Binary descriptors: BRIEF sampling, ORB orientation and learning, Hamming distance
- 3HOG: cell histograms, block normalization, and rigid-template matching
- 4Match scoring with L2 and Hamming distance, Lowe's ratio test, and cross-check
- 5Failure analysis: repetitive texture, low contrast, viewpoint change, and wrong-nearest-neighbor errors
Learning Outcomes
- Construct and normalize SIFT, ORB, and HOG descriptors
- Select an appropriate distance and apply ratio and cross-check tests
- Diagnose common matching failures from image and descriptor properties
Lecture Infographics
1 image file
Lecture Infographics
Lecture PDFs
Descriptors and Matching
The embedded PDF preview is disabled to reduce Vercel bandwidth usage. Open the lecture file directly in Google Drive for viewing, downloading, or presenting.