ENST/UOB/LU@TRECVID2007 HIGH LEVEL FEATURE EXTRACTION USING 2-LEVEL PIECEWISE GMM

摘要: We describe a high level feature extraction system for video. Video sequences are modeled using Gaussian Mixture Models. have used those models in the past to segment video into 2D+time objects. The segmentation result has been with great success compression scheme. In present work, components of model considered completely corresponding objects Their parameters as low-level features high-level detection topic or feature. is not optimized particular and thus scalable any number features. A threshold manually selected each after normalization. only difference between runs was tested two runs: B_ENST_1 uses znorm per B_ENST_2 use second provided better results. This an initial that will be explore effectiveness modeling videos GMMs.