Semantic indexing of a video document is a process
that performs the identification of elementary and complex
semantic units in the indexed document in order to create a
semantic index defined as a mapping of semantic units into the
sequences of video frames. Semantic content-based video retrieval
system is a software system that uses a semantic index built over
a collection of video documents to retrieve the sequences of video
frames that satisfy the given conditions.
This work introduces a new multilevel view of data for the
semantic content-based video retrieval systems. At the topmost
level, we define an abstract view of data and we express it in a
notation of enhanced conceptual modeling suitable for the formal
representation of the semantic contents of video documents.
A semistructured data model is proposed for the middle level
representation of data. At the bottom level we implement a
semistructured data model as an object-relational database. The
completeness of the proposed approach is demonstrated through
the mappings of a conceptual level into a semistructured level
and into an object-relational organization of data. The paper
describes a system of operations on semistructured data and
shows how a sample query can be represented as an expression
built from the operations.