【摘要】：直播電視產(chǎn)品已經(jīng)廣泛的走入了人們的日常生活中,如CNTV現(xiàn)已擁有央視直播、衛(wèi)視直播、城市直播、數(shù)字直播,共140余路高清直播,并且數(shù)目還在不斷增加,全國各省級、市級、地級的數(shù)字直播頻道,已經(jīng)逐步被完全覆蓋。直播電視主要由不同種類的電視節(jié)目構(gòu)成,如新聞資訊類、電視談話類、文藝類、娛樂類、記錄片類等等。能夠運用計算機設(shè)備,實時高效的判定各種電視節(jié)目的具體類別,這種技術(shù)將是非常具有科研意義和實際應(yīng)用價值的。廣告實時檢測系統(tǒng)選取的切入點是直播電視節(jié)目,通過對廣告片段判定算法進(jìn)行深入細(xì)致的研究,用基于學(xué)習(xí)的視頻類別判定技術(shù)和切合實際的應(yīng)用程序架構(gòu),來實時的分析和標(biāo)注直播電視節(jié)目中的廣告片段,并實現(xiàn)了從視頻采集到結(jié)果展示的完整系統(tǒng),省去了人工標(biāo)注視頻資料的耗費,使視頻分類算法轉(zhuǎn)化成了生產(chǎn)力�；趦�(nèi)容的視頻類別判定技術(shù)是多媒體研究領(lǐng)域中的一個重要課題。首先,基于內(nèi)容的視頻判定技術(shù)可以運用于日益增加的視頻數(shù)據(jù)管理、分類儲存和視頻內(nèi)容檢索上；其次,基于內(nèi)容的視頻類別判定技術(shù)可以有效的管理網(wǎng)絡(luò)視頻媒體的海量視頻,自動維護(hù)更新數(shù)據(jù)庫的標(biāo)簽,支持用戶對視頻的檢索與查找；最后,基于內(nèi)容的視頻類別判定技術(shù)可以用在不良視頻審查以及國家控制內(nèi)容審查上,對巨量的視頻數(shù)據(jù)中的違規(guī)違禁內(nèi)容進(jìn)行篩選、剔除或報警。本系統(tǒng)實現(xiàn)了基于內(nèi)容的廣告視頻類別實時標(biāo)注與判定算法,能達(dá)到在線運行演示效果的效率。本文研究了使用視頻內(nèi)容特征做視頻分類的技術(shù),針對直播電視中廣告視頻的判定,采用真實的CNTV數(shù)據(jù)作實驗,取得到了良好的效果,實現(xiàn)了完整的實時的在線識別廣告的系統(tǒng)。利用廣告鏡頭的尾幀(EFC)特性,提取具有代表性的圖像特征,采用SVM分類的方法區(qū)分廣告尾幀；結(jié)合廣告時長的統(tǒng)計特征作為后處理閾值判斷,劃分廣告片段；并且對算法做了很多工程上的優(yōu)化,能實時的展現(xiàn)識別效果。本文的主要特點是：該系統(tǒng)有著很高的廣告邊界定位精度,識別的廣告尾部和實際的廣告尾部極為接近；廣告片段識別準(zhǔn)確率較高,可達(dá)到83%以上；并且算法效率很高,使得系統(tǒng)可以實時地運行在CNTV在線視頻上。
[Abstract]:Live TV products have been widely into people's daily life, such as CNTV now has CCTV live broadcast, satellite TV live broadcast, urban live broadcast, digital live broadcast, a total of more than 140th high-definition live broadcast, and the number is still increasing, the provincial, municipal and prefectural digital live broadcast channels have been gradually fully covered. Live TV is mainly composed of different kinds of TV programs, such as news and information, TV conversation, literature and art, entertainment, documentary and so on. It is of great scientific research significance and practical application value to be able to use computer equipment to determine the specific categories of various TV programs in real time and efficiently. The starting point of advertising real-time detection system is live TV program. Through the thorough and detailed research on the algorithm of advertising segment determination, the video category determination technology based on learning and the practical application program architecture are used to analyze and mark the advertisement fragment in live TV program in real time, and the complete system from video capture to result display is realized, which saves the cost of manually marking video data. The video classification algorithm is transformed into productivity. Content-based video category determination technology is an important topic in the field of multimedia research. Firstly, content-based video decision technology can be used in increasing video data management, classification storage and video content retrieval. Secondly, content-based video category decision technology can effectively manage the massive video of network video media, automatically maintain and update the label of the database, and support users to search and find video. Finally, content-based video category determination technology can be used in bad video review and state-controlled content review to screen, eliminate or alarm a large number of illegal and illegal content in video data. This system realizes the real-time tagging and judging algorithm of advertising video category based on content, and can achieve the efficiency of running demonstration effect online. In this paper, the technology of video classification using video content features is studied. aiming at the determination of advertising video in live TV, the real CNTV data is used as an experiment, and good results are obtained, and a complete real-time online identification advertising system is realized. Using the (EFC) characteristics of the tail frame of the advertising lens, the representative image features are extracted, and the SVM classification method is used to distinguish the tail frame of the advertisement. Combined with the statistical features of the advertising time length as the post-processing threshold to judge, the advertising fragments are divided. And a lot of engineering optimization of the algorithm is made, which can show the recognition effect in real time. The main characteristics of this paper are as follows: the system has high accuracy of advertising boundary location, the recognition of advertising tail is very close to the actual advertising tail; the accuracy of advertising segment recognition is high, up to 83%; and the algorithm efficiency is very high, so that the system can run on CNTV online video in real time.
【學(xué)位授予單位】：復(fù)旦大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TP18;TN949.1

【共引文獻(xiàn)】

相關(guān)碩士學(xué)位論文 前1條

1 劉淑榮;基于語義的視頻檢索關(guān)鍵技術(shù)研究[D];華北電力大學(xué);2013年

，

本文編號：2499457