基于多特征組合和SVM的視頻內容自動分類算法研究
發(fā)布時間：2018-10-25 13:45
【摘要】： 基于內容的視頻自動分類是多媒體分析領域中一個重要的研究課題,它為日益增加的視頻數(shù)據(jù)的管理提供了方便,基于內容的視頻自動分類作為視頻傳播控制的一類關鍵技術在對網(wǎng)絡媒體進行有序管理的需求下至關重要。基于視頻自動分類技術的應用,媒體網(wǎng)站可以把海量的視頻內容進行自動分類,可實現(xiàn)對不良視頻信息的自動初步篩選。并且它在VOD和智能HDTV的發(fā)展中也發(fā)揮著重要的作用。 基于內容的視頻分類性能極大地依賴于視頻特征的提取和分類模型的選取,本文從對視頻內容和視頻風格類型的角度出發(fā),提出了基于視覺多特征組合的視頻特征提取方法和改進支持向量機(SVM)視頻分類算法,實現(xiàn)了對卡通、廣告、音樂、新聞和體育這五類最常見的視頻自動分類。 首先,在分析現(xiàn)有的視頻分類算法的基礎上,針對現(xiàn)有算法存在的問題,通過分析五類典型視頻在視覺上的差異,本文提出了新的特征表達方案即多特征組合模型,從編輯、顏色、紋理、運動四個方面提取了共十五種特征來構成新的視覺特征組合模型,所選用的特征空間增強了同類樣本分布的緊致性和異類樣本分布的差異性。在有效性和區(qū)分度上達到了滿意的效果。 在選擇并提取了合適的特征后,針對目前統(tǒng)計方法中存在的通過小樣本集很難設計有效分類器的問題,本文提出了基于支持向量機的視頻內容自動分類算法。并對分類器判決策略方法進行了改進,提出了基于動態(tài)閾值邊界向量抽取方法的一對多決策方法;基于二次預測機制的一對一決策方法;和基于交叉驗證概率誤差檢驗機制的有向無環(huán)圖決策方法。 通過仿真實驗結果說明:本文算法在特征選擇方面增強了五類視頻的區(qū)分能力,而且降低了單一特征的計算復雜度;其次,提高了SVM分類器的多視頻分類的性能;最后,與相關算法進行了對比實驗,證明了本文算法在分類正確率方面性能最佳。
[Abstract]:Content-based video automatic classification is an important research topic in the field of multimedia analysis, which provides convenience for the increasing management of video data. As a kind of key technology of video propagation control, content-based video automatic classification is very important under the demand of orderly management of network media. Based on the application of automatic video classification technology, media websites can automatically classify the mass of video content, and realize the automatic preliminary screening of bad video information. And it also plays an important role in the development of VOD and intelligent HDTV. The performance of content-based video classification greatly depends on the extraction of video features and the selection of classification models. This paper presents a video feature extraction method based on visual multi-feature combination and an improved support vector machine (SVM) video classification algorithm, which realizes the automatic classification of cartoon, advertisement, music, news and sports. First of all, based on the analysis of the existing video classification algorithms, aiming at the problems existing in the existing algorithms, by analyzing the visual differences of five kinds of typical video, this paper proposes a new feature representation scheme, namely, multi-feature combination model. Fifteen features are extracted from color, texture and motion to form a new visual feature combination model. The selected feature space enhances the compactness of similar sample distribution and the difference of heterogeneous sample distribution. Satisfactory results have been achieved in terms of effectiveness and differentiation. After selecting and extracting suitable features, aiming at the problem that it is difficult to design effective classifier by small sample set in current statistical methods, this paper proposes an automatic video content classification algorithm based on support vector machine (SVM). A one-to-many decision method based on dynamic threshold boundary vector extraction method and a one-to-one decision method based on quadratic prediction mechanism are proposed. And the decision making method of directed acyclic graph based on cross-validation probabilistic error test mechanism. The simulation results show that the algorithm enhances the ability of distinguishing five kinds of video in feature selection, and reduces the computational complexity of single feature. Secondly, it improves the performance of multi-video classification of SVM classifier. The experimental results show that the proposed algorithm has the best performance in classification accuracy.
【學位授予單位】：上海交通大學
【學位級別】：碩士
【學位授予年份】：2009
【分類號】：TP391.41
本文編號：2293871
[Abstract]:Content-based video automatic classification is an important research topic in the field of multimedia analysis, which provides convenience for the increasing management of video data. As a kind of key technology of video propagation control, content-based video automatic classification is very important under the demand of orderly management of network media. Based on the application of automatic video classification technology, media websites can automatically classify the mass of video content, and realize the automatic preliminary screening of bad video information. And it also plays an important role in the development of VOD and intelligent HDTV. The performance of content-based video classification greatly depends on the extraction of video features and the selection of classification models. This paper presents a video feature extraction method based on visual multi-feature combination and an improved support vector machine (SVM) video classification algorithm, which realizes the automatic classification of cartoon, advertisement, music, news and sports. First of all, based on the analysis of the existing video classification algorithms, aiming at the problems existing in the existing algorithms, by analyzing the visual differences of five kinds of typical video, this paper proposes a new feature representation scheme, namely, multi-feature combination model. Fifteen features are extracted from color, texture and motion to form a new visual feature combination model. The selected feature space enhances the compactness of similar sample distribution and the difference of heterogeneous sample distribution. Satisfactory results have been achieved in terms of effectiveness and differentiation. After selecting and extracting suitable features, aiming at the problem that it is difficult to design effective classifier by small sample set in current statistical methods, this paper proposes an automatic video content classification algorithm based on support vector machine (SVM). A one-to-many decision method based on dynamic threshold boundary vector extraction method and a one-to-one decision method based on quadratic prediction mechanism are proposed. And the decision making method of directed acyclic graph based on cross-validation probabilistic error test mechanism. The simulation results show that the algorithm enhances the ability of distinguishing five kinds of video in feature selection, and reduces the computational complexity of single feature. Secondly, it improves the performance of multi-video classification of SVM classifier. The experimental results show that the proposed algorithm has the best performance in classification accuracy.
【學位授予單位】：上海交通大學
【學位級別】：碩士
【學位授予年份】：2009
【分類號】：TP391.41
【引證文獻】
相關碩士學位論文 前2條
1 莫詠柳;基于支持向量機的聯(lián)機手寫漢字識別的研究[D];太原理工大學;2011年
2 趙宇熙;一種小型AUV的控制系統(tǒng)研究[D];哈爾濱工程大學;2012年
，本文編號：2293871
本文鏈接：http://www.wukwdryxk.cn/wenyilunwen/guanggaoshejilunwen/2293871.html
