| With the development of multimedia technology and network technology, video resources are becoming richer and richer. It is of vital importance to browse the video efficiently and retrieve a video clip automatically from huge amount of data. The main method to solve these problems is summary and retrieval of video and the important step is the segmentation of video. Compared with other video, news video, typical as it is, consists of a series of news stories and thus has more obvious structural features. Accurate detection and segmentation of the news stories play an important role in automatic search of news video and semantic understanding.In order to detect and segment news stories, segmentation methods based on anchorperson identification or audio-visual features and text information are currently used, which produce a low accuracy rate of about 85%. In view of the problems existed in the segmentation of news stories, this paper proceeds from two perspectives: detection of news caption and segmentation of news stories combined with news caption detection.(1) Captions of news video have rich edge information and significant features of lines. Based on the above characteristics, this paper proposes a news captions detection method based on edge detection and line features. On test with different news video frames of different channels in our experiments the detection recall rate about 93% and the detection accuracy rate about 94% are obtained. Results show that this algorithm has high detection recall rate and high detection accuracy rate.(2) According to the lower accuracy rate of news story segmentation, this paper proposes an automatic segmentation method of news video based on multi-feature fusion. Detection and segmentation of news stories are carried out by making use of the features such as mute, shot cut, anchorperson characteristics and text information. On test with different news video in our experiments the average segmentation accuracy rate about 93% is obtained. |
PDF Full Text Request