Maximum a posteriori adaptation method for video semantic indexing

To manage large amount of video data, an effective search mechanism is necessary. The keyword based search system is not efficient for video data due to the lack of metadata; hence for video indexing a method called Maximum-a-posteriori (MAP) method which uses Expectation Maximization algorithm to form a universal background model (UBM) by applying all training data. MAP adaptation uses a prior knowledge of UBM model parameters to estimate parameters of every training and test data. GMM Supervectors can be generated from the adaptive mean vectors. Support Vector Machine (SVM) along with GMM supervectors is used for the classification of video. Experimental evaluation of the proposed method is done in TRECVID 2010 video dataset and the result shows that it is better, since the method uses the fusion of visual and audio features.