The use of temporal, semantic and visual partitioning model for efficient near-duplicate keyframe detection in large scale news corpus.