| With the great development of the internet and modern multimedia service,digital video services have been integrated into people's lives and became an integral part of everyday life,the uploading and downloading of videos has also became an important source of getting information.Massive video has generated huge storage pressure,and it is an urgent problem that how to store,browse and retrieve these large-capacity video data reasonably,the video summarization technology just can alleviate this pressure.Therefore,video summarization technology has became one of the most popular research directions in the field of computer vision.It is widely used in various scenes in security work and video data analysis in smart cities.In the research of video summarization generation technology,the extraction of key frames is one of the core tasks.In the traditional algorithm,only the color information of the image is considered.However,only relying on the color feature can not reflect the main information of an image,and also can not reduce the interference caused by image noise,affecting the quality of key frames and reducing the effect of video summarization.In order to extract representative key frames and generate video summaries that match the user's viewing,this paper proposed a video summarization generation technology based on feature fusion,and research the video summarization generated by the classification algorithm and the clustering algorithm.Firstly,the paper researched the application of classification algorithm based on feature fusion in video summarization,and proposed a key frarme extraction method based on HOG-LBP feature and SVM classifier.This method uses Support Vector Machines(SVM)as a classification tool,extract and fuse shape and texture features from the decomposed video sequence,classify the detected images with the trained SVM classifier,and classify the correct frames as a candidate keyframes.The image cosine similarity method is used to eliminate redundant frames,and the quality of key frames is further improved.The exp erimental results show that the proposed algorithm extracts key frames with accurate average rate of 94.08%and the average error rate is only 23.18%.The extracted key frames have higher accuracy and representative content,and the effect is better.Secondly,in view of the fact that the classification algorithm is limited by the type and quantity of training samples,the shortcomings of the scene can only be roughly classified.The paper analyzed and learned the method of combining multiple features and clustering algorithm to select key frames of video summarization.The paper proposed an algorithm based on feature clustering to extract the key frame of video summarization.Firstly,the color features and texture features are selected and merged,and the initial clustering parameters are obtained by the hierarchical clustering algorithm.Then,the initial clustering result is optimized by K-means algorithm,and the frame closest to the cluster center is selected as the key frames;Finally,the effectiveness of the algorithm is evaluated by an objective and subjective ways.The experimental results show that the proposed algorithm extracts key frames with average precision of 0.71 in four different videos,the average recall rate is 0.76,and the average F-score reaches 0.73,which is compared with the current mainstream algorithm for extracting key frames.In terms of subjective analysis,the invited 10 users have higher recognition of the key frames generated by the algorithm,and some of the summary videos can reach a good level.Finally,form a video summarization based on all key frames extracted by different algorithms is according to time order that appears in each original video,which is played in the system working interface and obtain basic information of the summarization.The video summarization is more in line with the original content in chronological order,and it is more plot-developing and more suitable for the user's viewing habits. |
PDF Full Text Request