Dear Fernando, Frank, Yong, Rob, everybody,

I am sorry I have been away from the discussion for some time, due to travels and local reasons. However I read all the mails on the reflector. It seems we are progressing in the questions which are asked, and in probably in some answers !

This is a comment on Fernando proposals about interoperability, followed by comments on features and description schemes.

1. QUERY ENGINES INTEROPERABILITY

I think that there are several types of interoperability :

* applying queries on different description schemes. I agree that querying should be possible by any query engine on any description scheme. This is one of the goals of MPEG-7 standardization.

An example of a query engines which can apply to any object types is OQL, in the object and object database context.

* interpreting descriptors in another description scheme, or description schemes conversion. This issue is addressed in several research areas :

*  defining a standard query language. On this topic, we must pay attention to several issues : * Defining retrieval methods. I think that this is also a very delicate issue, as it might mean to standardize the query engine.
  However we might be careful about more extensive language standardisation, though any research progress in this language area would clearly be useful for most of us.

2. FEATURES AND DESCRIPTION SCHEMES

About features and description schemes, the  group may have done some progress. I am still thinking about what we could contribute to this discussion. I shall send you some other comments later. However, here are a few ones :

* As well as Fernando suggested, I believe that it is *very important* that features may support "subjective" as well as "objective" features. As long as agents do not have subjective beliefs (an probably further !) I don't think that this type of features may be extracted automatically (and not necessarily "semi-automatically"). However there are a lot of issues to gather and exploit such subjective features (see works in the W3C context).

For example I believe Franck suggested that a feature may be the "surealistic aspect of a scene (or object)". However, I don't believe that *for the moment*, there are automatic tools to extract "surealistic", or still more "beautiful" or "romantic". These are just examples.

* From a subjective feature point of view, and partly for "objective" content description (e.g. this is a balloon, a car, etc.)  I don't think that relevant features should all be pre-defined in the standard. This is a major reason to have extendible descriptors and description schemes.

Therefore the way from features to description schemes is not only defining the specific semantic of a given feature, but also specifying the list of features, except if features are defined in a very general way such as "content description" (the general goal of MPEG-7 !).

Probably we have done a significant progress by having now non-circular definitions, where features are defined separately from description schemes. Maybe we could go a step further and also try and independent definition of description schemes.

Some attempts by Yong Rui are quite interesting. In his mail, he proposes to define a description scheme as a model characterized by :

O = O(D, F, R), where D are the data, F, the features, R the representations. We could make this model more complete.
 

Therefore we would evolve to the following model type :

O = O (D, F, R, S, A, V)
 

We could notice that each feature may come with a (virtual) "similarity-retrieve" method, but that this method should be a slot for any implementation, it should not be part of the standard, and possibly might even not be part of the description scheme (?).

Yong, everybody, what do you think of it ?

Any comment ?

Regards & Season Greetings,

Pascal

-------------------------

Pascal Faudemay
LIP6 / UPMC (University Paris 6, Paris, France)

Pascal.Faudemay@lip6.fr        (at lab)
pascalf@club-internet.fr           (at home and when abroad)
tel / fax :                                   +33 01 44 27 71 16

--------------------------

Fernando Pereira wrote:

Lisboa, 23 December 1997

 Dear Rob, Frank, all

 According to the MPEG-7 Context and Objectives document, it is clear
that MPEG-7 wants to address both the push and pull models of content
identification (filtering versus retrieval).
The sentence "The question of finding content is not restricted to
database retrieval applications; also in other areas similar questions
exist. For instance, there is an increasing amount of (digital)
broadcast channels available, and this makes it harder to select the
broadcast channel (radio or TV) that is potentially interesting." is a
clear indication of this double approach.

 Moreover the MPEG-7 Context and Objectives and the MPEG-7 Requirements
documents state that
"MPEG-7, formally called "Multimedia Content Description Interface",
will standardize:
· A set of description schemes and descriptors, and
· A way to specify description schemes and descriptors.
MPEG-7 will address the coding of these descriptors and description
schemes."

 My big question here is:

 Is it enough to standardize descriptors and description schemes as well
as ways to specify additional descriptors and description schemes to
achieve an acceptable level of interoperability both for the push and
pull models ?

 I think this is a very important question since it is related to what
will effectively be MPEG-7 (the normative elements) with a clear impact
on the future call for proposals and consequent collaborative work.

 My personal problem is related to the pull model since I believe the
push model is well covered with the normative elements defined above.

 While in the push model the user will receive the descriptions and will
locally filter the relevant data using whatever type of querying and
matching criteria, in the pull model the user does not have direct
access to the descriptions which may be located in several databases
around the world. In the pull model, the descriptions reside somewhere
and are the queries that 'travel' to the descriptions.
 So imagine the case where I am at my home and I want to retrieve
information in as many as possible MPEG-7 enabled databases around the
world. Suppose that these databases already use the MPEG-7 normative
descriptions as stated above.

 IS THIS ENOUGH TO ACHIEVE AN ACCEPTABLE LEVEL OF INTEROPERABILITY ?

 How do I express my queries without loosing interoperability this means
without restricting the retrieval to the few databases that understand
my queries although they all use MPEG-7 descriptions ?

1) DONT'T WE NEED FOR THUS A MPEG-7 STANDARDIZED QUERY LANGUAGE ?

 This would allow all the MPEG-7 enabled databases to understand for
sure my queries because they would be expressed in MPEG-7 language,
increasing the interoperability.

 And what about the query matching criteria ? Will the user always be
obliged to get the data which matches the databases criteria or will it
be possible to the user to also express his particular query matching
criteria ?

2)  DON'T WE NEED FOR THIS A MPEG-7 STANDARDIZED WAY OF EXPRESSING QUERY
MATCHING CRITERIA ?

 This would allow all the MPEG-7 enabled databases to understand for
sure my query matching criteria because they would be expressed in a
MPEG-7 standardized way, increasing the interoperability.

 I hope somebody will help me with these very basic questions because
otherwise we will have to conclude that we need to change something very
important: the list of MPEG-7 normative elements.

 Season's Greeting for you and family

                                        Fernando

--
Fernando Manuel Bernardo Pereira, Ph.D., Professor
Instituto Superior Técnico,  Secção de Telecomunicações
Av. Rovisco Pais, 1096 Lisboa Cdx, PORTUGAL
Phone:  + 351 1 8418460           Fax:    + 351 1 8418472
E-mail: Fernando.Pereira@lx.it.pt WWW: http://amalia.ist.utl.pt/~fp/