Following these considerations, our approach is based on four main propositions.
1. Definition of an interchangeable audiovisual TV news description schema,
called NewsSchema. NewsSchema allows the representation of different news
video structure and content interpretations. It is based on MPEG-7, which
is the new standard for multimedia content description. We believe that MPEG-7
is an adequate means to be used for news video description, thanks to its
Data Definition Language (DDL) and a set of descriptors (Ds) and Description
Schemes (DSs). The DDL is based on XML schema, which offers several capabilities
such as defining data types and object oriented modelling. As for Ds and DSs,
they allow describing different aspects of the content, such as colour, temporal
and spatial features. We have studied the adaptability of these Ds and DSs
for the news video description and we have proposed some improvements to MPEG-7
in order to support the features needed in TV news applications. Based on
this study, we proposed the NewsSchema.
2. Definition of an audiovisual news query language, called NewsQL.
News video documents contain rich content information. The query language
should take into account different possible interpretations of the news video
content, and provide various types of access to the content based these interpretations.
In order to provide this possibility we have defined a set of views, each
one insisting on different aspects of the news video. One of the important
aspects of the news video, which is present in different views, is the structure.
Our query language should provide the facility of describing structural aspects
of the video to access the content. Another important feature is to provide
approximate matches and the ranking of the results in the style of information
retrieval systems. The language should be as extensible as possible in order
to take into account new descriptions of the content and also other types
of structured video documents.
3. Integration of querying and browsing. Besides querying, browsing
is another useful mode for accessing video content information. Browsing and
querying are indeed two complementary retrieval strategies whose mutual advantages
and drawbacks are quite complementary: in terms of orientation and cognitive
load problems (being disadvantages of browsing versus querying) and in terms
of free access to documents (being advantage of browsing versus querying).
Moreover, we consider that in the special case of video documents, the hierarchical
inter-document browsing is also a very effective way to verify the content
of videos resulted from querying. This capability has two advantages: first,
it transforms the time consuming linear method of video content verification
into a rapid hierarchical method. Second, it allows verifying for each video
segment its more general context (by going up the tree) and more details of
it (by going down the tree). This is specially needed in order to provide
a flexible way of exploring the database content and also to let the users
get familiar with the video structure and description scheme. Browsing inside
the structure of a news program is also a rapid way of content verification
after the querying phase. We believe that an integration of querying and browsing
can be very helpful to facilitate the retrieval of complex multimedia documents
and specially in our case of news video documents.
4. Customisable news retrieval interface. The definition of a platform,
which specifies how the special features of different news video types (represented
in the NewsSchema) influence the creation of querying and browsing interfaces.
We have currently implemented a querying and browsing interface based on the
TSR NewsSchema. These interfaces represent the special logical structure of
the TSR news. In order to provide a system capable of querying different types
of news video it is very helpful to have such platform, which defines how
to reflect the structural aspects of news video in querying and browsing interfaces.
The two first issues cited above are the main concerns, which are currently
sconsidered in the PhD research of Nastaran Fatemi.