Research On Key Techniques Of Video Retrieval Based On Ontology

In recent years,with the rapid development of science and technology,video-based multimedia technology has also risen rapidly,and a large number of growing video data have entered daily life.Finding the video you need from a huge amount of video has become a hot topic of concern.The traditional text-based retrieval not only has the problem of wasting time and consuming a lot of manpower,but also affects the retrieval efficiency because of people's subjective consciousness,resulting in the inability to accurately retrieve the required video or retrieve a large number of redundant video.Compared with the former,content-based video retrieval has relatively improved the precision and recall of video retrieval,but still needs further improvement.This paper first proposes video shot boundary detection algorithm based on HSV and mutual information.Secondly,it proposes a key frame extraction method based on custom K-value clustering and content analysis.Finally,a labeling algorithm combining probability association and ontology is proposed.Key technologies such as shot segmentation,key frame extraction and annotation improve the precision and recall of video retrieval.Efficient video retrieval is based on accurate annotation of video information.Video files that are too large,too long and too much redundant information can cause situations that are difficult to accurately label.Therefore,before the video is annotated,the complete video is accurately segmented into a single independent lens,and then extracts key frames from the single shot to remove redundant information and marks them with key frames of the shot,finally achieving the video marking effect.The low precision and recall rate of video retrieval is mainly caused by problems such as lighting flash,object motion being susceptible to the surrounding environment,and misunderstanding caused by artificial semantic information.Therefore,this paper first proposes a lens edge detection algorithm based on HSV and mutual information to reduce the impact of illumination flash and object movement.Secondly,this paper proposes a key frame extraction method based on custom K-value clustering and content analysis to extract more accurate key frames and lay a foundation for accurate video annotation.Finally,a probabilistic association labeling algorithm combined with ontology is proposed to effectively eliminate semantic ambiguity and break the semantic gap.Experiments,show that it is feasible to use the method proposed in this paper for shot segmentation,key frame extraction and video annotation,and one of the prominent features of the algorithm is that it can effectively find the required video clips in a large number of redundant and complex video libraries and improve the precision and recall of the video required by people to retrieve.