Introduction - If you have any usage issues, please Google them yourself
This paper presents a technique for combining global context with local SIFT information to produce a feature
descriptor that is robust to local appearance ambiguity and non-rigid transformations