The Research Of Shot Boundary Detection Algorithm Based On Content

With the rapid growth in multimedia information, the volume of digital video data has increased significantly.Huge amount of digital data pose a challenges in many digital video analysis domains: video storage,video compression, video indexing and video content retrieval. Therefore, this has created a need for tools that can efficiently organize video data into more compact forms. Layer management is very convenient for video analysis.Video shot boundary detection is the basic and critical step in video structure analysis. video shot boundary detection technology was studied in this paper. Respectively aiming at problems in the missing detection and false detection,efficient solution algorithms were given in this paper.In order to solve the problem of missing detection, an efficient algorithm for shot boundary detection is proposed in this paper.In order to eliminate the disturbances caused by illumination and camera motion,Dual Tree Complex Wavelet Transform(DTCWT) is used in the process of raw feature generation.Then,the enhanced texture features were extracted from the six detailed subbands by calculate the mean and standard deviation of the high-frequency coefficient at each level. Non-negative matrix factorization(NMF) is used for extracting structure feature from the low-frequency sub-bands image.And dynamic threshold is used for shot boundaries detection.This method experiments on TRECVID 2001 test data and other test videos shows that the proposed scheme can overcome illumination and motion effects efficiently and achieve a high detection speed.Considering the problem of false detection,in this paper,we propose a two-stage framework to detect video shot boundary.At the first stage, Dual-tree complex wavelet transform(DT-CWT) is used to eliminate the disturbances caused by illumination change or fast camera and object motion.Then,we use non-negative matrix factorization(NMF) instead of traditional metrics to extract structure feature and employ adaptive threshold to initial detect shot boundary.Owing to large amounts of data in the video, NMF can significantly decreased the computing load and the time delay.Video shot boundary redetection algorithm based on motion estimation and compensation is proposed in the second stage by eliminating the error detection in the result of initial detection.The proposed method has been validated on TRECVID 2001 test data and other test videos,experimental results show that the proposed scheme can overcome illumination and rapid motion effect efficiently and achieve a high detection speed.