Analysis, Based On The Detection And Extraction Of The News Video Subtitles

The newscast is one of the ways for people to obtain the information and it plays animportant role in people's daily life. The detection and retrieval based on newscast'scaption is an important portion of the content-based video information classification andretrieval, which is the main topic in the multimedia information research.This paper chooses news videos as its research object and focuses its study on thedetection and contraction of the three common types of video caption, mainly including,video shot detection, topic-caption detection, caption line split, caption binary, charactersegmentation and recognition, etc.As to the video structure, based on analyzing of recent researches, the paper achievesthe video structure with the texture consistency analysis of video spatio-temporal slice.As to the caption detection, this paper offers the caption detection method based ontexture features of spatio-temporal slice in shorts, having realized the detection of the twocommon types of non-rolling caption region boundary. It avoids detecting all the captionframes so as to reduce the complexity of detection. Based on the texture character ofrolling caption in spatio-temporal slice, it detects the rolling caption with up and down,offering prototype system of caption detecting, which achieves good experiment results.As to the caption retrieval, having analyzed the disadvantages of the two references,this paper puts forward the caption location method based on involving wavelet weightedreconstruction and morphology. It employs the thrice Interpolation method to enhance thecaption. It compares several commonly used binaries algorithm and adopts Otsu algorithmto binary caption. It uses the projection algorithm to partition characters. At the same time,it analyzes the common problems of projection algorithms while partitioning the charactersand offers some solutions. Finally, it uses the given Han Wang OCR software to recognizethe characters.