UEC at TRECVID 2008 High Level Feature Task.
Zhiyuan Tang, Akitsugu Noguchi, Keiji Yanai · 2008
In this paper, we describe our approach and results for high-level feature extraction task (HLF) at TRECVID2008. This year, our focus is to develop a framework which fuses a number of features effectively. In our paper, color, face, motion, text, and local pattern features were extracted. After that, a simplymodified version of Adaboost algorithm was implemented as a late fusion to combine all these features. Description of our submitted runs is as follows: • (Run1)UEC fusion ver6, (Run2)UEC fusion ver2, (Run3)UEC fusion ver1, (Run4)UEC fusion ver5: fusion of color, face, motion, text and Bag-of-Features (BoF) model of local pattern features by using a simple version of Adaboost. use different setting to compute error rate in the fusion phase. • (Run5)UEC fusion c bdd ver6, (Run6)UEC fusion c bdd ver2: fusion of color, face, BoF model of local pattern features by using a simple version of Adaboost. use different setting to compute error rate in the fusion phase. Run1∼Run4 are the same to combine color, face, motion, local pattern features by using our algorithm. In our experiment, we changed some parameters when computing the error measure in the fusion algorithm, this makes the 4 runs different from each other. By the analysis of the results of these 4 runs, we noticed that motion and text did not help us at all, so we also tried to fuse only color, face, local pattern features as Run5 and Run6 with different error measure computation. As a result, Run5 yielded the best performance (infAP=0.0314) of these our 6 runs. 1.