A hybrid domain enhanced framework for video retargeting with spatial–temporal importance and 3D grid optimization

• Domain-specific knowledge is incorporated to allocate resources to user preferred contents.
• Visual attention, semantic analysis, and temporal consistency are combined to generate a spatial–temporal importance map.
• Spatial–temporal 3D rectilinear grid framework is proposed in an optimal manner.
• The framework easily adapts to any video domain.

Recently, a ubiquitous video access is highly demanded for online video applications. One big challenge is that video service needs to adapt different device capabilities. Pervasive multimedia devices require an accurate and user comfort video retargeting. Letting users see their preferred content accurately directly affects their comforts. User preferences on video contents are different in various video domains. In this paper, we present a hybrid framework of video retargeting with a domain enhanced spatial–temporal grid optimization. First, we parse videos from low-level features to high-level visual concepts, combining with visual attention for an accurate importance description. Second, a semantic importance map is built up representing the spatial importance and temporal continuity, which is incorporated with a 3D rectilinear grid scaleplate to map frames to a target display, thereby keeping the aspect ratio of semantically salient objects as well as the perceptual coherency. Extensive evaluations are made on five typical video genres, i.e. sports, advertisements, lecture, news and surveillance. The comparison with the state-of-the-art approaches on both images and videos have demonstrated the advantages of the proposed approach.