Font Size: a A A

Research On Video Shot Boundary Detection And Keyframe Extraction Technology

Posted on:2020-05-22Degree:MasterType:Thesis
Country:ChinaCandidate:Y WangFull Text:PDF
GTID:2428330575967960Subject:Computer technology
Abstract/Summary:PDF Full Text Request
The development of information technology has produced a large amount of video data in different application fields.Content-based video retrieval technology has become an important issue in recent years due to the difficulty in organizing,indexing,and retrieving video data.Content-based video retrieval technology mainly researches techniques such as shot boundary detection of video,keyframe extraction of video,and video retrieval.This paper is a research and practice based on the current research status of video retrieval in the context of electronic innovation in the publishing industry,which requires effective storage,indexing and retrieval of audio and video files.The research work of this paper has the following points:On the basis of learning the traditional video shot boundary detection based on color and texture features,combined with the advanced nature of current depth learning technology in image recognition and image feature extraction,the existing shot boundary detection method is improved.After combining the convolutional neural network features with the traditional color histogram features,the threshold parameters are adjusted experimentally,and the accuracy of the abrupt shot boundary detection is improved compared with the traditional method.Based on the existing gradual shot boundary detection method,a gradual shot boundary detection method based on recurrent neural network is proposed.Firstly,a network model suitable for classifying video frames is designed.By collecting a certain amount of training data,the network model is trained and verified,and finally it can complete the classification task of recognizing the gradual transition frame and the normal frame with higher accuracy.On the basis of learning the traditional keyframe extraction method based on frame difference and the keyframe extraction method based on clustering,according to the steps of k-means clustering,combined with the temporal characteristics of video frames,the clustering method is improved and the results of the class are sequential.Finally,combined with the actual project,design a video retrieval experiment and a video similarity measurement method that can meet the experimental requirements.Through experimental tests,the video sea rch has a high accuracy rate when searching for similar videos,which can meet the actual project requirements.
Keywords/Search Tags:video shot boundary detection, recurrent neural network, keyframe extraction, k-means clustering, video retrieval
PDF Full Text Request
Related items