广告视频检索技术研究

发布时间：2018-05-16 12:41

本文选题：镜头边界检测 + 广告单元边界检测　；参考：《南京理工大学》2013年硕士论文

【摘要】：随着视频数据的快速增长,如何对视频进行有效地管理、存储和检索成为用户十分关注的问题。本文以广告视频为研究对象,对广告视频检索相关技术进行了研究,设计并实现了一个广告单元检索系统。对视频进行了结构化的处理,采用基于运动补偿的自适应双阈值方法来判定突变镜头和渐变镜头。在此基础上,针对电视视频流中的广告板块,对广告板块边界检测的算法进行了改进。首先提取并计算镜头特征作为基本单元,然后采用概率输出的SVM模型对视频进行分类,最后在搜索窗口内搜索概率极值判断出极大似然切分点。对分割出的广告板块,利用小波变换等提取图像特征,对其进行滑动窗口分割,判断出镜头组边界信息。对广告板块中的音频,提取底层音频特征如Mel频率倒谱系数等,采用基于GMM和分割熵的检测算法,实现了广告板块中音频跳变检测。在总结了现有的广告单元分割方法的基础上,利用镜头组边界检测结果、音频跳变的检测结果和镜头边界检测结果,设计了一种基于音视频融合的广告单元分割方法,实现了广告单元边界检测。在广告视频的检索方法上,提出一种新的基于奇异值分解的SURF-DAISY关键帧匹配方法,对相似度模型的实际应用进行了研究,并提出了广告视频检索的具体步骤。本文采用Visual C++为开发平台,实现了一个广告视频检索系统,验证了上述方法。
[Abstract]:With the rapid growth of video data, how to effectively manage, store and retrieve video has become a very important issue for users. In this paper, the advertising video retrieval technology is studied, and an advertising unit retrieval system is designed and implemented. The video is processed structurally, and the adaptive double threshold method based on motion compensation is used to determine the abrupt shot and the gradient shot. On the basis of this, the algorithm of advertising block boundary detection in TV video stream is improved. Firstly the shot feature is extracted and calculated as the basic unit then the SVM model of probability output is used to classify the video. Finally the maximum likelihood segmentation point is determined by searching the maximum probability extremum in the search window. The image features are extracted by wavelet transform, and the moving window is segmented to judge the boundary information of shot group. The bottom audio features such as Mel frequency cepstrum coefficients are extracted for the audio in the advertising block. The detection algorithm based on GMM and segmentation entropy is used to realize the detection of audio jump in the advertising block. On the basis of summarizing the existing methods of advertising unit segmentation, using the results of shot group boundary detection, audio jump detection and shot boundary detection, a method of advertising unit segmentation based on audio and video fusion is designed. The advertisement unit boundary detection is realized. In this paper, a new SURF-DAISY key frame matching method based on singular value decomposition (SVD) is proposed. The practical application of similarity model is studied, and the concrete steps of advertising video retrieval are proposed. In this paper, an advertising video retrieval system based on Visual C is implemented, and the above method is verified.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41

【参考文献】