Abstract
The paper brings a novel method for generating visual summaries of scalable videos. The generated summaries can dynamically adapt to requirements defined by display size, userpsilas needs or channel limitations. It utilises compressed domain features coupled with efficient contour evolution algorithm in order to generate a scale space of temporal video descriptors. The layout of the visual summary is created using an efficient graph clustering technique and a fast discrete optimisation algorithm, enabling dynamic video summarisation in real-time. The experimental results show good scalability of the dynamic layout and highly efficient generation of visual summaries.