| The explosive growth of online video has made it increasingly urgent to browse and manage interesting video content quickly and accurately.Video summarization is an effective technique to solve this problem by converting one or more videos to a compact summarization.Different from single video summarization,multiple videos are quite noisy,redundant,and even irrelevant,which makes it a challenging task to find important representative visual content from this complex video information.The traditional multi-video summarization method does not produce satisfactory results because it ignores the user's search intent.To this end,we first propose a novel Query-Aware Sparse Coding method for Multi-Video Summarization(QUASC),which provides comprehensive and concise information about the query videos.Specifically,it designs a novel query-aware approach by formulating the multi-video summarization in a sparse coding framework,where the web images searched by the query are taken as the important preference information to reveal the query intent.In order to provide a user-friendly abstract,we present an event keyframe structure by using unsupervised multi-graph fusion method to detect the specific event group related to the query and presenting the keyframe with this structure.Secondly,we propose a Multi-Video Summarization method with Importance-Aware Sparse Auto-Encoder(MVS-IASAE),which adds query-based web images as a constraint of importance to sparse auto-encoder to guide the acquisition of important video content.The summarization of the importance,representativeness and diversity are jontly considered.In addition,a bottom-up ranking algorithm is proposed to present summarization to improve the readability.Finally,we construct a large multi-video summarization dataset called MVS1 K.The proposed QUASC and MVS-IASAE algorithms are experimentally validated on the MVS1 K dataset.Compared with several advanced multi-video summarization methods,the proposed is effective and advanced. |
PDF Full Text Request