Fig. 10From: COGNIMUSE: a multimodal video database annotated with saliency, events, semantics and emotion with application to summarizationIllustration of saliency curves, summarization algorithm, and emotion curves. Monomodal saliency curves for the three modalities (audio, visual, and text). Multimodal saliency, av vs. avt, where the weight w=0.1 and summarization algorithm (illustrating the effect of reconstruction opening by showing the raw (automatically) selected segments, the shot reconstructed and the speech reconstructed segments) and the manual selected segments (by the human annotators). Emotion annotation illustrated by the two dimensions arousal and valence; for 3000 frames, scene from the movie “CRA”. Keyframes (top) correspond to the indicated saliency peaks. Best viewed in colorBack to article page