The ever-increasing spatial/temporal resolution of video such as ultra high definition has raised new challenges on storage, transmission and display for business, home, and mobile video applications, and thus efficient representation, compression, and retargeting of high-resolution (HR) videos become the key issues for effective deployment of these applications. Saliency models can facilitate to address these issues, but in practice, the current state-of-the-art saliency models are insufficient for efficiently handling complicated scenes containing multi-scale non-homogenous objects, highly textured regions and cluttered background.The objective of this project is to propose an efficient spatiotemporal saliency model to predict salient regions in HR videos, and fully exploit it to ease the design and improve the performance of HR video compression and retargeting applications. With the aim to overcome the drawbacks of existing saliency models, based on a multiscale region representation, the proposed model systematically realizes statistical model saliency measuring, intra-scale saliency modification, inter-scale saliency propagation and flexible incorporation of top-down information, to generate a novel saliency representation form with scalability, saliency tree, from which a multiscale saliency fusion scheme is used to derive high-quality saliency maps at various scales. Saliency tree enables an efficient search of multiple salient objects, guides the temporal non-uniform downsampling, and directs the enhanced mode decision in HR video compression. Saliency diffusion, shrinkability/stretchability estimation, and an integration of cropping/warping and uniform scaling are exploited for HR video retargeting.The research results will consolidate Europe as a leader in the research domain of saliency modeling and saliency based applications, facilitate to promote the developments of HR video services in Europe, and thus boost the European competitiveness.