| Faced with massive and complex video structures,how to quickly and accurately retrieve video content of interest has become a hot and difficult research topic.At present,the traditional text-based retrieval method has been difficult to fit the new requirements,so content-based video retrieval came into being.Content-based video retrieval is a retrieval method that matches feature similarity by studying the features of image frames.It mainly includes shot boundary detection,key frame extraction and other technologies.In this paper,shot boundary detection and key frame extraction are studied.The main work is as follows:(1)A feature extraction algorithm based on HSV color histogram and GLCM texture feature fusion is proposed.Non-uniform block extraction of HSV color features and GLCM texture features,appropriate weight allocation,70% of the weight allocated to the color component(HSV),30% of the weight allocated to the texture component(GLCM),in order to fuse features to characterize the image frame.(2)A shot boundary detection algorithm based on multi-feature fusion is proposed.The image frame features are extracted and the similarity index between frames is calculated.The shot boundary is judged by combining global threshold with adaptive double threshold.In this way,the randomness of artificial threshold setting is avoided,and the confusion of gradient,mutation or flash is effectively distinguished.The accuracy and recall rate of shot boundary detection are improved.(3)A new key frame extraction algorithm based on K-Means++ clustering and comparison is proposed.K-Means++ clustering is used to extract the key frames initially,and then the key frames output from the clustering are compared with each other to remove the nearly duplicate key frames.The extracted key frames not only maintain the integrity of shot content,but also reduce the redundancy between key frames.(4)The proposed algorithm is analyzed through experiments to prove the effectiveness of the proposed algorithm. |
PDF Full Text Request