Abstract. Concept indexing in multimedia libraries is very useful for users searching and browsing but it is a very challenging research problem as well. Beyond the systems ’ implementations issues, semantic indexing is strongly dependent upon the size and quality of the training examples. In this paper, we describe the collaborative annotation system used to annotate the High Level Features (HLF) in the development set of TRECVID 2007. This system is web-based and takes advantage of Active Learning approach. We show that Active Learning allows simultaneously getting the most useful information from the partial annotation and significantly reducing the annotation effort per participant relatively to previous collaborative annotations. 1

This paper presents the systems used by CLIPS-IMAG to perform the Shot Boundary Detection (SBD) task, the Feature Extraction (FE) and the Search (S) task of the Video track of the TREC-11 conference. Results obtained for the TREC-11 evaluation are presented.

This paper presents the systems used by CLIPSIMAG to perform the Shot Boundary Detection (SBD) task and the Feature Extraction (FE) task of the TRECvid workshop. Results obtained for the 2003 evaluation are presented. The CLIPS SBD system based on image dierence with motion compensation and direct dissolve detection was second among 14 systems. This system gives control of the silence to noise ratio over a wide range of values and for an equal value of noise and silence (or recall and precision), the value is 12 % for all types of transitions. Detection of person X from speaker recognition alone was deceiving due to the small number of shots containing person X in the overall test collection (about 1/700) and the even small number in which person X was actually speaking (about 1/6000). Detection of person X from speech transcription performed much better but was still lower than other systems using also the image track for the detection.

Concept indexing in multimedia libraries is very useful for users searching and browsing but it is a very challenging research problem as well. Beyond the systems’ implementations issues, semantic indexing is strongly dependant upon the size and quality of the training examples. In this paper, we describe the collaborative annotation system used to annotate the High Level Features (HLF) in the development set of TRECVID 2007. This system is web-based and takes advantage of Active Learning approach. We show that Active Learning allows simultaneously getting the most useful information form the partial annotation and significantly reducing the annotation effort per participant relatively to previous collaborative annotations. 1

In this paper, we compare active learning strategies for indexing concepts in video shots. Active learning is simulated using subsets of a fully annotated dataset instead of actually calling for user intervention. Training is done using the collaborative annotation of 39 concepts of the TRECVID 2005 campaign. Performance is measured on the 20 concepts selected for the TRECVID 2006 concept detection task. The simulation allows exploring the effect of several parameters: the strategy, the annotated fraction of the dataset, the number of iterations and the relative difficulty of concepts. Three strategies were compared. The first two respectively select the most probable and the most uncertain samples. The third one is a random choice. For easy concepts, the “most probable ” strategy is the best one when less than 15% of the dataset is annotated and the “most uncertain ” strategy is the best one when 15 % or more of the dataset is annotated. The “most probable ” and “most uncertain ” strategies are roughly equivalent for moderately difficult and difficult concepts. In all cases, the maximum performance is reached when 12 to 15 % of the whole dataset is annotated. 1.

This paper presents the system used by CLIPS-IMAG to perform the Shot Boundary Detection (SBD) task of the Video track of the TREC-10 conference. Cut detection is performed by computing image dierence after motion compensation. Dissolve detection is performed by the comparison of the norm over the whole image of the rst and second temporal derivatives.

The aim of our investigation was to explore a new method of analysing flow images, based on the Optical Flow Technique. Conventionally, this technique was developed for detecting motion of large objects in a real world scene. Applied to the flow images, it appears to be an interesting alternative offering high evaluation accuracy without most of the typical limitations characteristic of FFT based PIV. Besides evaluation of tracer images, the new method was also tested with smoke images obtained from experiments both in a fluidised bed and in a wind tunnel. It was also successfully tested on an image sequence of a vapor bubble growing on a thin heated wire. The accuracy of the velocity measurements using the new implementation was investigated using synthetic particle image sequences generated with the help of a 2D numerical simulation. 1 Introduction The aim of this investigation is to explore the possibility of using an optical flow technique in measuring fluid flow velocity. Classic...

The demo presents an optical flow computationtechnique based on dynamic programming, its results on the Fleet, Barron and Beauchemin benchmark, and its application to: 3D reconstruction, time interpolation of image sequences using motion information, particle image velocimetry and segmentation of video sequences into shots. 1.

CONTENTS 1 Contents 1 Introduction 3 2 Probl'ematique 4 2.1 Architecture des machines informatiques . . . . . . . . . . . . . . . . . . . . . . . 4 2.2 Exp'erimentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 2.3 Sp'ecificit'es de la perception . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 3 Les processeurs de reconnaissance des formes 8 3.1 Le coprocesseur de programmation dynamique . . . . . . . . . . . . . . . . . . . 9 3.2 Le microprocesseur de Comparaison Dynamique (PCD) . . . . . . . . . . . . . 12 3.3 Le syst`eme multiprocesseur de reconnaissance phon'etique . . . . . . . . . . . . . 14 3.4 Bilan et perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 4 Le Calculateur Fonctionnel 18 4.1 Le mod`ele d'ex'ecution flots de donn'ees cabl'es . . . . . . . . . . . . . . . . . . . . 19 4.2 La programmation fonctionnelle . . . .

This paper presents a system that performs the recovery of camera motion parameters and the segmentation of mobile objects in video documents for content indexing. Two dierent methods are used for the recovery of the camera motion (relatively to the main background), the rst for a camera maintained at a xed location with rotational and zoom degrees of freedom, and the second for a camera of arbitrary motion but assuming a xed focal length. The rst method is based on the search of an optimal projective transform between consecutive images combined with an iterative background / mobile objects segmentation process. The second method is based on a paraperspective factorization method for shape and motion recovery. Both methods rely on the use of a dense and high-quality matching between consecutive images (optical ow). The system also attempts to classify shots or sub-segments of shots into one of the following categories of \no motion", \non mobile camera motion", \mobile camera motion" or \other type of motion"