Crossing the gap between multimedia data and semantics

Crossing the gap between multimedia data and semantics DBDM 2007 - ASSIGNMENT 2 JASPER A. VISSER (9306137) BART J. WITTEMAN (1100688) EYAL HALM (0324329)

Data and semantics Data is of no value without context Multimedia data often lacks this context, needs semantics for Accessibility Usability Data mining

The problems of multimedia semantics Data is high volume Data has high complexity Data is heterogeneous Therefore: multimedia semantics are hard to assign

Crossing the semantic gap Different approaches are possible: Extract features from media automatically (AI): Data-driven crossing Adapt and expand the data Expert couples semantics Include matching semantics in query: User-driven crossing Adapt the query

Data-driven semantics Add metadata to multimedia Use expert knowledge to train automated systems Requires linkage of multimedia to metadata Metadata should be descriptive of actual data Large amounts of data should be analyzed

User-driven semantics Change the way the user queries the data Example: picture selection using relevance feedback Requires appropriate feature sets

State of the art Semantic Retrieval of Multimedia Data: http: //oswinds. csd. auth. gr/papers/mmdb 04. pdf

State of the art Learning Semantic Categories for 3 D Model Retrieval http: //portal. acm. org/c itation. cfm? id=129008 2. 1290090

State of the art Combining several approaches to support query processing and relaxation: Supporting efficient multimedia database exploration: http: //www. springerlink. com/c ontent/x 4 bcw 95 tnx 8 pfn 90/

Further considerations Temporal data: e. g. still versus moving pictures Level of detail tradeoffs Feature selection

Questions and discussion