Boris Mansencal

Rushes summarization by IRIM consortium: redundancy removal and multi-feature fusion

In this paper, we present the first participation of a consortium of French laboratories, IRIM, to the TRECVID 2008 BBC Rushes Summarization task. Our approach resorts to video skimming. We propose two methods to reduce redundancy, as rushes include several takes of scenes. We also take into account low and mid-level semantic features in an ad-hoc fusion method in order to retain only significant content

Clustering of scene repeats for essential rushes preview

This paper focuses on a specific type of unedited video content, called rushes, which are used for movie editing and usually present a high-level of redundancy. Our goal is to automatically extract a summarized preview, where redundant material is diminished without discarding any important event. To achieve this, rushes content has been first analysed and modeled. Then different clustering techniques on shot key-frames are presented and compared in order to choose the best representative segments to enter the preview. Experiments performed on TRECVID data are evaluated by computing the mutual information between the obtained results and a manually annotated ground-truth.

The COST292 experimental framework for rushes summarization task in TRECVID 2008

In this paper, the method used for Rushes Summarization task by the COST 292 consortium is reported. The approach proposed this year differs significantly from the one proposed in the previous years because of the introduction of new processing steps, like repetition detection in scenes. The method starts with junk frames removal and follows with clustering and scene detection; then for each scene, repetitions are detected in order to extract once the real scene; the following step consists in face detections (faces are considered semantically relevant) and in pan, tilt and zoom detections (other camera motions are usually related to technical operations in the backstage); finally the summary is extracted.

A multi-object particle filter tracking with a dual consistency check: Application to mid-level concept detection in videos

A novel mid-level video indexing method based on detection and tracking human faces is presented. Instead of detecting the faces on every frame, our method first detects the faces and then tracks them. Compared to our previous general-purpose tracking method, our approach is improved by: i) a Multi-Object model extension to track several objects in parallel; ii) a Dual Consistency Check by Kolmogrov-Smirnov test to alarm a scene change so as to stop the tracking and wait until the next detection; ii) application of temporal median filtering of initial detection by Viola & Jones detector. The combination of filtered detection and our tracking method evaluated on an excerpt of TRECVID 2009 database increases the F-measure by 7% compared to Viola & Jones detector alone.