Section: Software
Software
Several software programs have been developed in the team over the years: I-Description (APP deposit number: IDDN.FR.001.270047.000.S.P.2003.000.21000 ), Asares (IDDN.FR.001.0032.000.S.C.2005.000.20900) , AnaMorpho (IDDN.FR.001.050022.000.S.P.2008.000.20900) , Faestos (IDDN.FR.001.470029.000.S.P.2006.000.40000) , 2pac (IDDN.FR.001.470028.000.S.P.2006.000.40000) , DiVATex (IDDN.FR.001.320006.000.S.P.2006.000.40000), NaviTex (IDDN.FR.001.190034.000.S.P.2007.000.40000) and Telemex . Please refer to the previous activity reports for a description of these programs. In 2008, the following pieces of software were developed or updated.
Vidsig
Participants : Sébastien Campion [ correspondent ] , Patrick Gros.
The application vidsig , developed in the framework of Xavier Naturel's thesis, enables to compute a video signature of various video formats (TV and Web) Based on Discrete Cosinus Transformation, it produces video signature small and robust (64 bits per image).
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
Vidseg
Participants : Sébastien Campion [ correspondent ] , Patrick Gros.
The vidseg software, developed in the framework of Manolis Delakis thesis, allows to detect video segment in various formats. Results are composed by significant segmentation features such as silences in audio track, changes of ratio aspect, monochrome images, cuts and dissolves.
Vidseg was registered at APP (IDDN.FR.001.250009.000.S.P.2009.000.40000)
Isec
Participant : Sébastien Campion [ correspondent ] .
Isec (I mage S earching E ngines C omparator) is a web application used as graphical interface for image searching engines based on retrieval by content. It was developed with the help of Pierre Cosquer during an internship.
Its architecture allows to plug different searching engines and enables to compare their results. It also permits to edit submitted images with several operations (cropping, blurring, etc.). These attacks make it possible to test the robustness of searching engines.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
GPU-KMeans
Participants : Nguyen Khang Pham, Sébastien Campion [ correspondent ] , François Poulet.
GPU-KMeans is an implementation of k-means algorithm on graphical process unit (graphic cards) using CuBLAS library. These hardware properties (data-parallel computing) allow GPU-KMeans to reduce compute operations by a factor 10.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
Correspondence Analysis
Participants : Nguyen Khang Pham, Annie Morin, Sébastien Campion [ correspondent ] .
Correspondence Analysis software is designed to compute a factorial correspondence analysis (FCA) for image retrieval. FCA analyzes a contingency table crossing terms/words and documents. To adapt FCA on images, Correspondence Analysis uses "visual words" computed from scalable invariant feature transform (SIFT) descriptors in images and use them for image quantization.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
GPU Correspondence Analysis
Participants : Nguyen Khang Pham, Sébastien Campion [ correspondent ] .
GPU Correspondence Analysis is an implementation of the previous software Correspondence Analysis on graphical processing unit (graphical card) which reduces execution time of compute operations by a factor 20.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
CAVIZ
Participants : Nguyen Khang Pham, Annie Morin, Sébastien Campion [ correspondent ] .
CAVIZ (C orrespondance A nalysis VIZ ualisation ) is an interactive graphical tool, which allows to display and to extract knowledge from the results of a Correspondence Analysis on images.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
Kiwi
Participants : Gwénolé Lecorvé, Pascale Sébillot [ correspondent ] .
Kiwi (standing for Keyw ords E xtrator) is a software which seeks to convert textual documents into weighted word vectors using the TF-IDF method, widely used in information retrieval. Hence, Kiwi is mostly dedicated to indexing and keyword extraction purposes. Besides raw texts, the standard TF-IDF method has been extended to process transcriptions and word meshes, to return word n-uple vectors, and to integrate external knowledge on words such as confidence measure or mutual information.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
Topic Segmenter
Participants : Camille Guinaudeau, Pascale Sébillot [ correspondent ] .
In collaboration with G. Gravier from the Metiss project-team.
The topic-segmenter program is a software dedicated to topic segmentation of texts and (automatic) transcripts, mostly based on lexical cohesion, implementing (and extending) a method described in the paper from Masao Utiyama and Hitoshi Isahara, "A Statistical Model for Domain-Independent Text Segmentation", ACL, 491–498, 2001.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
S2E
Participants : Sébastien Campion [ correspondent ] , Mathieu Ben.
S2E (S tructuring E vents Extractor) is a Python module which allows the automatic discovery of audiovisual structuring events in videos. The algorithm is based on a cross-modal mutual information measure between audio and video clusters. These mono-modal clusters are obtained from two separate hierarchical clustering processes, one on audio segments, the other one on video shots. The module outputs a segmentation of the video with a score for each segment indicating how confident we are that a structuring event occurs during the corresponding time slot.
Registration at the “Agence pour la Protection des Programmes” (APP) is in progress.
