Research On Shot Segmentation And Key Frame Extraction Techniques In Video Retrieval

In recent years,with the rapid development of digital control technology and network technology,the amount data of video resources is growing explosively.And the rapid development of the internet makes it easier to transmit and share video.It is becoming more and more important to store and manage these video data.All of these situations have created tremendous demand for video retrieval technology and have motivated many scholars to further advance this technology research.Video retrieval is applied to get the video clips in the vast video data.The traditional method is to annotate the key words for video by manual annotation.This text-based method is simple but subjective,and time-consuming.So it cannot be used as an effective management and retrieval method of video.The content-based video retrieval method analyzes and processes the information characteristics of the video frames,and then adaptively selects the desired video by the related algorithm.This method can more flexibly and efficiently store,manage and retrieve video data,this technology has begun to attract more and more attention.In the video retrieval,how to accurately segment the shot of video is the basis for improving retrieval accuracy.How to effectively extract the representative frame from video is an important step to improve the performance of retrieval system.In this paper,the existing problems and the advantages of shots segmentation and key frame extraction in video retrieval are analyzed and arranged.On the basis of previous studies,this paper proposes corresponding improvements.The main research is as follows:(1)In study of shot segmentation,a video shot segmentation method is proposed based on dual-detection.It consists of two processes: pre-detection and re-detection.At pre-detection stage,the video frames are divided into the target and background.Features of target are extracted by Hu moment invariant and detect the dissimilarity;at re-detection stage,the three-dimensional color histogram is used to further detect the results of first stage and get the shot boundary;According to the different transition time of the abrupt and gradual of shots,the sliding window is used to detect the transition time of shots.In order to prove the accuracy and stability of the algorithm,four different types of video data are segmented in experiment of this paper.Compared with other segmentation algorithms,the recall and precision rate of this method are all above 90% and the rate of missed detection is low.The efficiency of the overall algorithm is better than other methods.(2)In study of key frame extraction,a key frame extraction method is proposed based on Discrete Cosine Transform(DCT)and Nonlinear Correlation Information Entropy(NCIE).In order to maintain the image quality and reduce the amount of computation,firstly,the corresponding coefficients are extracted by DCT to replace the low-level feature of video frame.According to the complexity of shots,the discrete coefficient is used to distinguish between content-rich shots and content-single shots to complete the preliminary classification.Secondly,NCIE metric method is used to measure the correlation of multiple frames in dynamic shots as a whole.The shots is further subdivided into several sub-shots.In the end,the most representative frame of the sub-shot and the middle frame of the static shot are extracted as key frames.Experiments have shown that compared with other methods,this method can improve the fidelity by 0.7% while reducing false detection rate and missed detection rate.It is also superior to other key frame extraction algorithms in subjective evaluation.