Overview
Syllabus
SIFT: David Lowe, UBC
SIFT - Key Point Extraction
Advantages
Invariant Local Features
Steps for Extracting Key Points
Scale Space (Witkin, IJCAI 1983) • Apply whole spectrum of scales
Approximation of LOG by Difference of Gaussians
Building a Scale Space
How many scales per octave?
Initial value of sigma
Scale Space Peak Detection
Key Point Localization
Initial Outlier Rejection
Further Outlier Rejection
Orientation Assignment
Similarity to IT cortex
Extraction of Local Image Descriptors at Key Points
Descriptor Regions (n by n)
Key point matching
Taught by
UCF CRCV