Research On Video Retrieval Technology Based On Multi-feature Fusion

With the development of self-media,the video content is rich and colorful,and the traditional features can no longer meet the needs of modern society for video retrieval.With the popularity of short videos,personal small videos are widely spread in the media,and the data volume of short videos is increasing rapidly.People pay more and more attention to the content they are interested in,hoping to find their favorite videos efficiently and retrieve the results they want.With the gradual maturity of deep learning technology,it is no longer difficult to train a particular interesting content.However,it is still a difficult point to efficiently retrieve videos of interested target people from massive data.Therefore,the research and practice are combined with the related technologies in deep learning.The specific contents of this paper are as follows:Most methods of extracting key frames by K-Means clustering do not consider the information of faces in images.Therefore,in order to quickly retrieve the video clips of the target characters,a key frame extraction method based on face recognition and improved K-Means clustering is proposed on the basis of K-Means clustering and related methods in face recognition.Firstly,face detection method is used to detect face pictures and extract face features.Then,K-Means clustering method is used to extract key frames.In K-Means clustering,the similarity boundary of faces is used to determine the clustering radius,and cluster classes are divided by boundary iteration.Finally,the face pictures closest to the center of clusters are taken as key frames.The final experimental results show that the accuracy of this method is improved compared with the methods in the literature.Traditional methods are used to extract features and fuse them,without considering the interested target person,but using a single feature such as the person's face or body posture has certain limitations on the accuracy.Therefore,this paper proposes a multi-feature fusion video retrieval method based on key frame extraction.Firstly,face features,head features and body posture features are extracted by using convolutional neural network,and these features are used as the standard to measure a person.Then,different features are used to match and fuse with the corresponding features in the video library.Finally,the video clips of the target people that users are interested in are retrieved.Experiments show that the retrieval method in this paper has higher accuracy than the literature method.To sum up,the experimental results show that this method can accurately extract key frames,and on this basis,retrieve videos more accurately.