| Recently,video data volume presents the explosive growth.How to quickly find content to meet the needs of users in massive video data has become the focus of attention.Unlike the low-level features,semantic reveals the deep meaning of the video, such as emotion and the links between objects.So the method of semantic based video retrieval is more closer to the way of human’understanding of the video.News video is a special kind of video which contains important practical information and closely linked with people’s life. Content based news video semantic concept detection technology can be used for video annotation and video retrieval, so as to improve the efficiency of video browsing,which is of great significance in various applications of news video.This paper mainly through the analysis of news video content to explore the underlying high-level semantics.In this paper, visual information, text information and audio information are comprehensively used to analyze news video. The primary technology involves video key frame extraction, video text extraction and video scene detection.My research contents include the following aspects:1. Video key frame extraction technology. Due to the characteristics of some shot switching points and mute points is consistent in news video, mute detection technology is combined to reduce the complexity of the subsequent processing. We firstly segment news video into several shots,and key frame extraction is based on lens change rate.Experiments show that this key frame extraction algorithm can achieve higher recall ratio and lower recurrence rate.2. Video subtitles extraction technology. Due to the particularity of news video:subtitles generally appear in the lower part of the video frames.We can only process the one fifth of the bottom area of the video frames so as to narrow the processing scope and improve the extraction rate.Corner detection method is used to detect text area,that is, through the statistics of angular point to judge whether the image contains text area.Morphology algorithm is used to enhance the text.Experiments show that this method is simple and effective.3. Image classification based news video scene detection technology.Classic Dense Sift features are extracted for the images and the current popular BOF algorithm is improved for feature mapping.This paper takes the probability density function gradient histogram statistics algorithm to improve the accuracy of the image description. SPM algorithm is used for feature integration to overcome the faults of losing space information of the traditional method.The classic SVM classifier is used and combined with Gram matrix algorithm to reduce the amount of computation of classification. |
PDF Full Text Request