本文選題：視頻內(nèi)容分析 + MPEG-2　； 參考：《北京工業(yè)大學(xué)》2016年碩士論文

【摘要】：視頻內(nèi)容分析(Video Content Analysis,VCA)一直是多媒體領(lǐng)域最受關(guān)注的研究課題之一,經(jīng)過多年的研究,取得了一定的研究進(jìn)展,目前已廣泛應(yīng)用于視頻近拷貝檢測、廣告投放以及視頻檢索等多個(gè)領(lǐng)域。視頻內(nèi)容分析的關(guān)鍵部分在于如何提取視頻特征,從而對視頻內(nèi)容進(jìn)行有效、全面的描述。近年來,圖像內(nèi)容分析技術(shù)、圖像局部特征、稀疏理論等不斷取得新的進(jìn)展。為此,本文將這些新的理論和方法應(yīng)用到視頻內(nèi)容分析中,深入研究了視頻內(nèi)容的特征提取與表征,然后將其應(yīng)用到視頻近拷貝檢測以及視頻廣告內(nèi)容的插入中。本文的研究內(nèi)容主要包括以下幾個(gè)部分:1.提出了一種面向MPEG-2視頻的時(shí)空特征提取與表達(dá)方法針對現(xiàn)有視頻內(nèi)容分析中提取的特征描述力不足的問題,結(jié)合視頻編碼格式的特點(diǎn),提取MPEG-2視頻的時(shí)間和空間特征充分表征視頻內(nèi)容。首先,利用視覺顯著性模型提取視頻關(guān)鍵幀,然后對關(guān)鍵幀提取HSV顏色直方圖、ORB特征(Oriented FAST and Rotated BRIEF)等空間特征,并對ORB特征進(jìn)行稀疏表示,同時(shí)自MPEG-2碼流中直接提取運(yùn)動(dòng)矢量,對其繪制角度直方圖為視頻的時(shí)間特征�？臻g特征和時(shí)間特征相結(jié)合,得到多維度的視頻時(shí)空特征,來表征視頻內(nèi)容。2.設(shè)計(jì)并實(shí)現(xiàn)了一種基于時(shí)空特征的MPEG-2視頻近拷貝檢測方法本文將提取的視頻時(shí)空特征表達(dá)應(yīng)用于視頻近拷貝檢測技術(shù)中。通過分別比較各個(gè)特征的相似度,利用基于投票機(jī)制的決策融合方法,綜合得出查詢視頻與參考視頻內(nèi)容相似度,從而給出近拷貝檢測結(jié)果。在標(biāo)準(zhǔn)數(shù)據(jù)集上的實(shí)驗(yàn)結(jié)果顯示,本文提取的時(shí)空特征可以有效抵抗多種近拷貝變化,同時(shí),提出的近拷貝檢測算法具有更優(yōu)的準(zhǔn)確率和檢測速度。3.設(shè)計(jì)了一種基于內(nèi)容的視頻廣告插入方法當(dāng)前的視頻廣告插入方法是在固定的時(shí)間點(diǎn)上將廣告插入目標(biāo)投放視頻,對視頻的播放過程造成嚴(yán)重干擾,極易引起瀏覽者對廣告商品的抵觸情緒。因此本文在現(xiàn)有的基于內(nèi)容的廣告插入方法基礎(chǔ)上,對其進(jìn)行了改進(jìn)。本文算法根據(jù)視頻的時(shí)空特征與結(jié)構(gòu)化特點(diǎn),計(jì)算廣告與投放視頻之間的內(nèi)容相似度,篩選出恰當(dāng)?shù)膹V告插入位置,實(shí)現(xiàn)了基于內(nèi)容的廣告插入。本文進(jìn)行了主觀評價(jià)實(shí)驗(yàn),實(shí)驗(yàn)結(jié)果表明,本文算法相較于定點(diǎn)插入方法,對瀏覽者的干擾較小。
[Abstract]:Video Content Analysis (VCA) has been one of the most concerned research topics in multimedia field. After many years of research, it has made some progress, and has been widely used in video near copy detection. Advertising and video retrieval and other areas. The key part of video content analysis is how to extract video features so as to describe video content effectively and comprehensively. In recent years, new advances have been made in image content analysis, image local features and sparse theory. Therefore, this paper applies these new theories and methods to video content analysis, studies the feature extraction and representation of video content, and then applies it to video near-copy detection and video advertising content insertion. The research content of this paper mainly includes the following several parts: 1. A spatio-temporal feature extraction and representation method for MPEG-2 video is proposed to solve the problem of insufficient description of features extracted from existing video content analysis, combined with the characteristics of video coding format. The temporal and spatial features of MPEG-2 video are extracted to fully represent the video content. Firstly, the video key-frame is extracted by visual saliency model, then the spatial features such as HSV color histogram Orb feature oriented FAST and Rotated BRIEF) are extracted from the key-frame, and the ORB features are represented sparsely, and the motion vector is extracted directly from the MPEG-2 bitstream. The angle histogram is the time feature of video. Combining spatial features with temporal features, a multi-dimensional video space-time feature is obtained to represent the video content. 2. This paper designs and implements a method of near copy detection for MPEG-2 video based on temporal and spatial features. In this paper, the extracted temporal and spatial feature representation of MPEG-2 video is applied to the near copy detection technology of video. By comparing the similarity of each feature and using the decision fusion method based on voting mechanism, the similarity of query video and reference video content is synthesized, and the result of near-copy detection is given. The experimental results on the standard data set show that the spatio-temporal features extracted in this paper can effectively resist a variety of near-copy changes. At the same time, the proposed near-copy detection algorithm has better accuracy and detection speed of .3. In this paper, a content-based video advertisement insertion method is designed. The current video advertisement insertion method is to insert the advertisement into the target video at a fixed point in time, which causes serious interference to the video playback process. It is easy to cause resistance to advertising products. Therefore, based on the existing content-based advertising insertion method, this paper improves it. According to the spatiotemporal and structural features of video, this algorithm calculates the content similarity between advertising and video delivery, selects the appropriate position of advertising insertion, and realizes content-based advertising insertion. The experimental results show that the proposed algorithm has less interference to the visitors than the fixed-point insertion method.
【學(xué)位授予單位】：北京工業(yè)大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2016
【分類號】：TP391.41
，

本文編號：1804620