| In recent years,with the rapid deve lopment of computer technology and multimedia technology,video and image resources are increasingly rich and g radually become the m ainstream media for inform ation exchange.How to obt ain the interest information from these massive amounts of videos and images is becoming a hot spot for multimedia technology researchers.With respect to the low-level inform ation such as color and lightness,the text inform ation in video and im age belongs to high-level semantic information,which is the important clue to understand the content of video and image.Extracting the text information is also of great significance to establish the content-based video/image retrieval system.In news video,the caption text is a hi gh-level semantic information,which is usually a summary of a news event and provi des an important clue to understand the content of the news video.Therefore,obt aining the caption text is helpful for the automatic annotation of the news video,which is also helpful for the establishm ent of the content based news video indexing system.This paper is m ainly for news video image,the relevant key technol ogy about text extraction is mainly researched,such as the text location algorithm and the character segmentation algorithm.The specific work as follows:(1)The caption text is usually with comp lex and changeable background in news video image,which m akes it d ifficult to ob tain the text inf ormation directly.So the location of caption text is an important prerequisite for obtaining the cap tion information.This paper is mainly for the caption text in news video image,based on the rich edge characteristics of text and combined with the characteristics of the caption text(location,size,etc.),a new m ethod based on Maximum Feature Score Region(MF SR)is proposed to extract the caption text.Compared with the traditional location algorithm based on projection histogram,the proposed algorithm has better results.In addition,the proposed method can also be utilized to extr act the dialogue subt itles in m ovies or teleplays images without language restriction.(2)In order to reduce the dif ficulty of character recognition and im prove the accuracy of the text recognition,the text line seg mentation algorithm is studied.In the case of caption line in news video im age(mainly Chinese characters),com bined with the structural characteristics of caption line,a text line segmentation algorithm based on template matching is proposed,which could quickly segment a text line into som e single characters.The proposed algorithm obtains the suitable segmentation paths based on the value of te mplate function,which could ef fectively overcome the problem that traditional algorithm could not accurately segment the Chinese characters. |
PDF Full Text Request