Video Shots Annotation using Random Forest

With dramatically increasing of video resources, manually semantic video annotation requires extensive human power. Automatic annotation is an efficient and appropriate solution. In this paper, a tag propagation scheme using random forest is applied on video shot semantic annotation. For the content representation of each video shot, multiple keyframes are extracted using K-means clustering method. We train random forest with tag distribution information gain criterion, and estimate the probabilities of assigning tags to annotate each keyframe. The final predicted semantic tags of video shot comes from the weighted summation of probabilities of assigning tags of all keyframes. The experimental results on videos indicate that our video shot annotation based on random forest achieves good performance.