Movie Keyframe Retrieval Based on Cross-Media Correlation Detection and Context Model

In this paper, we propose a novel cross-media correlation detection method for movie keyframe retrieval. We first compute the temporal saliency on both the video and audio streams in a movie separately, then locate the resonance regions that the saliency changes in these two modalities show strong correlations. Next, starting from resonance regions, we propagate the similarity of visual and auditory characteristics through neighboring movie regions based on a temporal movie context model, segmenting the movie into a sequence of coherent parts from which keyframes are extracted. The experimental results on actual movie clips show that, compared to the single-modality algorithms, our method gives improved retrieval performance in completeness and precision due to the efficient exploitation of the context and correlations between complementary multi-modalities.