Visual video retrieval using Multivariate GARCH models

This paper deals with a new video sequence retrieval that takes into account the evolution of features in time and the interactivity between objects in sequence. A low-level features extracted from key frames of video sequence are considered as observations in time series, which are modeled using Multivariate GARCH model (General Autoregressive Conditional Heteroscedasticity). Experiments are conducted with the framework of TRECVID, Sound and Vision 2007/8/9 video collection. Experimental results are presented to illustrate the good performance of the method.