For the more technically inclined.
Quote:
Digital video services become more and more popular nowadays. Huge video databases should be stored and processed with appropriate speed and quality. Many new problems are raised, such as video duplicates, search in video, copyright examination, video sorting, etc.
Video Matching Technology is designed to solve some of those problems. The main functionality of the technology is to find a part of a given video in a video database. There are two steps in the search process: 1) create index database and 2) find a video similar to a given sample using that index. The algorithm requires approximately 50 KBytes for index information in case of one and a half hours film. That leads to tractable index sizes even for large databases. Algorithm successfully works with both small half minute fragments and full-length films. Matching process is fully automatic. It is important to note that Video Matching Technology is robust to video deformations. Different types of video deformation are supported: brightness shift or equalization, hard noise, strong compression artifacts, resolution change, black borders (letter box) insertion or cropping, subtitle insertion, and many others. Search complexity is logarithmic with respect to database size. Video Matching Technology can work without original database by using only indexes. That can be used for remote video matching. A search result is a list of best matched videos from a database and matching probabilities of them which can be used for automatic matching decision.
Video Matching Technology can be used to find duplicates in video databases, build effective indexes of large video collections, for copyright examination.
|
It's nice that they can keep it in O(log n). That leads me to believe that they don't have to compare each fingerprint to every other fingerprint to find a match. They probably store it in a manner in which they specify the video generally and then narrow down more.
That would suggest that they have the ability to find similar videos as well.