News Video Semantic Mining Based on Multi-Wing Harmoniums Model

According to the heterogeneous features of news video accompany text and multimode key frame image,the Multi-Wing Harmoniums(MWH) model is set up,which contains two submodels of multivariate text Poisson distribution and multivariate color histogram Gaussian distribution. By researching the doubleaction dependent relation of implication theme and observation input,this model is extended to two layer random field model,which conducts the semantic mining to news video. It is tested in CCTV news video sets,and the results show this model is effective.