Team TEXMEX

Members
Overall Objectives
Scientific Foundations
Application Domains
Software
New Results
Contracts and Grants with Industry
Other Grants and Activities
Dissemination
Bibliography

Section: Software

Softwares

I-Description

Participant : Patrick Gros [ correspondant ] .

I-Description allows a user to compute local or global image descriptors: differential local invariants, global and local color histograms or weighed histograms. It was deposited with the ``Agence pour la Protection des Programmes'' (APP) under the number IDDN.FR.001.270047.000.S.P.2003.000.21000.

Asares

Participant : Vincent Claveau [ correspondant ] .

Asares is a symbolic machine learning system (based on inductive logic programming) that automatically infers, from descriptions of pairs of linguistic elements (noun-noun, noun-verb...) found in a corpus in which the components are linked by a given semantic relation (synonymy, hyperonymy, qualia, lexical function...), corpus-specific morpho-syntactic and semantic patterns that convey the target relation. The patterns are explanatory and linguistically motivated, and can be applied to a corpus to efficiently extract resources and populate semantic lexicons. Two semi-supervised versions of Asares also exist, that rely on a combination of the supervised symbolic pattern learner and a statistical extraction technique. These semi-supervised versions rival Asares 's supervised one. Asares was deposited with the ``Agence pour la Protection des Programmes'' (APP) under the number IDDN.FR.001.0032.000.S.C.2005.000.20900

Faestos

Participant : Pascale Sébillot [ correspondant ] .

Faestos (Fully Automatic Extraction of Sets of keywords for TOpic characterization and Spotting) is a tool composed of a sequence of statistical treatments that extracts from a morpho-syntactically tagged corpus sets of keywords that characterize the main topics that corpus deals with. The system exploits the distribution of words of the corpus over its paragraphs, and requires neither human intervention nor given knowledge about the number or nature of the topics of the corpus. The extracted lists of keywords are employed in order to detect the presence of a topic in a paragraph, revealed by a keyword cooccurrence. Morevover, the system extracts from each keyword class a triple of words that permits an intuitive designation of the underlying topic. Faestos was deposited with the ``Agence pour la Protection des Programmes''.

2PAC

Participant : Pascale Sébillot [ correspondant ] .

2pac (2-Pass Acquisition of semantic Classes) brings together words used in a similar way in a topical sub-corpus such as that extracted by Faestos so as to build classes of words of similar meanings ("semantic classes") specific to the use that is made of them in that given topic. It works by first computing general semantic proximities from the whole corpus, then using that information to perform a more in-depth analysis of one of the topical sub-corpora. The result is a classification tree from which classes must be extracted manually; a graphical user interface has been developed to ease that task of manual exploitation. 2pac was deposited with the ``Agence pour la Protection des Programmes''.

tmx-DIVA-solution

Participants : Cédric Dufouil, Arnaud Dupuis [ correspondant ] , Patrick Gros.

The library tmx-DIVA-solution (DIVA stands for Di stant V ideo A ccess) enables an easy access to a collection of videos stored on a remote machine. It contains a server on the remote machine, in charge of extracting from video files the image or sound data needed by the user and sending it to the client. The client part of the library decodes what it receives from the server and provides the user with the exact data needed (the exact image or sound information). A caching system allows the system to transfer a GOP only once when several neighbouring images are queried by a user. The server and the client can work on different systems (Windows, Linux or MacOSX). tmx-DIVA-solution was deposited with the ``Agence pour la Protection des Programmes'' (APP) under the number IDDN.FR.001.320006.000.S.P.2006.000.40000

NaviTex

Participants : Cédric Dufouil [ correspondant ] , Arnaud Dupuis, Patrick Gros.

The NaviTex tool was developed with the help of tmx-DiVA solution . It allows a user to index and annotate videos using TV-Anytime (XML) metadata format. Those metadata can be exported to the DiVA manager , in order to easily access some parts of a video directly from the web page. NaviTex runs on Linux and Windows OS. NaviTex is in the process of being deposited with the ``Agence pour la Protection des Programmes'' (APP).


previous
next

Logo Inria