Telefonica Research at TRECVID 2010 Content-Based Copy Detection
Ehsan Younessian, Xavier Anguera, Tomasz Adamek, Nuria Oliver · 2010
Abstract—This notebook paper presents the participation ofTelefonica Research in the task of Video Copy Detection inTRECVID 2010. This is our second participation and, for thisyear, we have developed two local-based monomodal systemsthat we then combine using a score-based fusion to obtain amultimodal system output. We submitted 4 runs in total, whosemain characteristics are described below: TID.m.[BALANCED/NOFA].fusion: These correspond toour main submission, both for the no false alarm andbalanced profiles. They are based on the fusion betweenthe local audio and local video monomodal systems. TID.m.BALANCED.videoonly: This submission is basedon the monomodal video-based system using DART localfeatures and with a temporal consistency postprocessing. TID.m.BALANCED.audioonly: This submission is based onthe monomodal audio-based system using frequency-basedaudio local features.From these four systems submitted, two of them are processingonly monomodal information (audio or video) and the fusionsystem takes the output of the previous two to output a fusedresult. Results for the monomodal systems in terms of NDCR arefar from optimal, mainly due to an exces of false alarms thatour monomodal systems still output. Results for F1 scores arevery good for all cases. When combining the monomodal systemsinto he fusion the NDCR scores improve quite a bit as most falsealarms are eliminated.The proposed fusion turned out to work very well for com-bining our two monomodal systems. We will further investigateit to improve it for future evaluations.