This thesis explores the motion anchoring strategies, which represent a fundamental change to the way motion is employed in a video compression system—from a “prediction-centric” point of view to a “physical” representation of the underlying motion of the scene. The proposed “reference-based” motion anchorings can support computationally efficient, high-quality temporal motion inference, which requires half as many coded motion fields as conventional codecs. This raises the prospect of achieving lower motion bitrates than the most advanced conventional techniques, while providing more temporally consistent and meaningful motion. The availability of temporally consistent motion can facilitate the efficient deployment of highly scalable video compression systems based on temporal lifting, where the feedback loop used in traditional codecs is replaced by a feedforward transform.The novel motion anchoring paradigm proposed in this thesis is well adapted to seamlessly supporting “features” beyond compressibility, including high scalability, accessibility, and “intrinsic” frame upsampling. These features are becoming ever more relevant as the way video is consumed continues to shift from the traditional broadcast scenario with predefined network and decoder constraints to interactive browsing of video content via heterogeneous networks.