Video key information extraction system aiming at user interest semantics

Abstract

The invention relates to a video key information extraction system aiming at user interest semantics. The system comprises an image acquisition module, a video preprocessing module, a joint team weight learning module and a key information extraction module. The method includes: taking interest description of different users on video as keywords to retrieve images on the Internet through the imageacquisition module; using the joint team weight learning module to retrieve the keywords which are different in description and related in semantics, and acquiring a target semantic model at the sametime; utilizing associated audio data of the video to assist pre-extraction of video clips probably containing key information; calculating whether each video clip contains the key information that users are interested in or not through the target semantic model. The system can perform semantic modeling on content that the users are interested in according to different concerns of the users so asto effectively guide personalized video key information extraction, and semantic knowledge acquired on the Internet can greatly reduce dependency of data on manual annotation.

Description

technical field [0001] The invention relates to a video key information extraction system aimed at user interest semantics, belonging to the field of graphics and image processing. Background technique [0002] Most of the traditional methods extract the highlights of the video based on the underlying features, ignoring the selection of the content that the user is interested in. Extracting video clips is a relatively subjective task. When the video content is complex or the duration is long, different users are interested in the content of the same video. For example, for a basketball game video, the content that users are interested in may be highlights such as slam dunks and three-pointers, or they may focus on technical actions such as players' ball-handling postures and court moves. The expectations for highlight extraction are different. [0003] In recent years, the research on extracting video highlights by using high-level semantics has gradually attracted the att...

AI Technical Summary

Problems solved by technology

However, videos on the Internet are usually randomly shot by non-professionals. The videos contain a lot of blur caused by jitter and intra-class transformations caused by complex shooting scenes. Direct semantic modeling of the video itself requires labeling a large number of videos. It is time-consuming and labor-intensive, and the blurring and shaking that may exist in the video will bring great difficulties to semantic modeling

Images