Subject: Lýsing á lokaverkefni
hefur verið samþykkt af lokaverkefnanefnd.
1 – INTRODUCTIONVideo retrieval: To query a database of videos whether it contains a copy of an input video clip, or possibly any similar clips. The input video may have been altered (modified or 'attacked'), and the database should still return meaningful results &ndash the system should be effective or robust. Querying should be efficient, i.e. fast, even with a large database.
Significant work has been done in the field of video retrieval to use local descriptors and various multidimensional [database indexing] methods. These have been shown to be rather effective in finding videos in large databases. The database side of things doesn't seem to be fully developed, compared with [certain indexing methods] that have proven successful in image retrieval. Furthermore, not many different kinds of video modifications have been evaluated in this type of system, modifications that [c certain type of local descriptor has been] shown to be resistant to in image retrieval.
The question is therefore whether [this combination of local descriptor and database indexing scheme is] applicable to content-based video retrieval, given that [it] has worked well in the quite similar field of image retrieval. A further question that we hope to answer is whether near-duplicate keyframe detection and removal can be used to usefully filter redundant information from videos prior to indexing and querying.