Séminaire de Melle Tran Thi Thanh Hai, chercheur permanente du Centre MICA - Date : jeudi 24 juin 2010, 16h00 - Lieu : Centre MICA
Intervenant : Melle Tran Thi Thanh Hai, chercheur permanente du groupe "images/vidéos et extraction sémantique", de retour d'un séjour de 2,5 mois au NII de Tokyo comme chercheur invité
Date : mardi 22 juin 2010, 15h00
Lieu : salle polyvalente, Centre MICA, Bat C10, IP de Hanoi (HUST)
Interprète traducteur : le séminaire aura lieu en français
Résumé/Abstract:
With the increasing of storage and digital devices, huge amounts of multimedia databases are available. This requires research of efficient methods for video analysis and indexing. The video analysis is to detect and recognize concepts of user interest from different modalities such as video stream, audio, text. A concept could be a static object, a landscape or an event.
Semantic concept video detection consists of representing video by a descriptor and compares this descriptor with concept models that have been learnt before. Two questions are which feature should be used, how they are combined together to efficiently represent video clip containing a certain concept?
Appearance features such as Color, Texture, SIFT, SUFT, LBP are good to represent static objects but they cannot be able to express movement of objects as well as the relation between objects in the scene.As the appearance based features cannot completely describe all aspects of the scene (what and how), the performance of appoaches based only on appearance features remains limited for event detection.
We think that motion is useful cue to represent event and help to distinguish them better. This work focuses then on studying motion feature; compare it with appearance feature and evaluate on a benchmark data TRECVID 2007.
