【作者】 王晓晨；

【导师】 武剑洁；

【作者基本信息】 华中科技大学 ， 软件工程， 2015， 硕士

【摘要】 随着计算机网络的普及和数字电子设备的广泛使用,视频已成为信息传播的主要载体之一。如何方便用户在海量的数据库中快速、准确地查找感兴趣的视频,逐渐成为人们关注的焦点。在此背景下,基于内容的视频检索算法逐渐成为研究热点。基于视觉词袋模型的视频检索属于基于内容的视频检索范畴,该算法主要包括两部分:第一部分是关键帧提取算法,第二部分是构建图像的加权视觉词袋模型算法。针对传统关键帧提取算法存在的冗余问题,提出了一种将图像颜色特征和边缘特征相结合的两阶段的关键帧提取算法。该算法充分利用图像的颜色特征和边缘特征,对比图像帧之间的相似度,选出最具代表性的关键帧,描述镜头的核心内容,构建关键帧数据库。在关键帧提取算法基础之上,构建图像的加权视觉词袋模型,实现视频的精确查询。首先提取图像局部特征,生成描述图像信息的高维向量,将计算图像之间的相似度转化为计算高维向量之间的距离。然后计算查询图像与视频之间的相似度,并按照相似度由高到低输出满足用户查询要求的视频。该算法自动提取视频特征,并描述视频内容,无需人工参与,从而避免了因人工标注不够精确而导致的检索失败。对视频提取关键帧,构建关键帧数据库,测试该算法的检索性能。实验验证该算法能够独立实现视频检索,检索返回结果准确、客观,具有较高的查全率和查准率。更多还原

【Abstract】 Video has become one of main carriers of information dissemination, with the widespread use of computer networks and the popularity of digital electronic devices.Much research focuses on how to find out interest video in massive database quickly and accurately.The method called “Content-Based Video retrieval(CBVR)” becomes a hotspot.The Algorithm called “ video retrieval algorithm based on bag of visual word model” belongs to CBVR. The proposed method consists of two parts: the first part is keyframe extraction algorithm, and the second part is a weighted bag of visual word model of image construction. A two-stage frame extraction algorithm is adopted to solve the redundancy problem in the traditional keyframe extraction method. The proposed method makes good use of color and counter features, compares the similarity between image frames, selects the most representative keyframe, improves the retrieval efficiency,and constructs key frame database.Image precise query is achieved by weighted bag of visual word model based on the keyframe extraction algorithm. The model generates a high-dimensional vector to describe the image information through local feature extraction. The location of keyframe information is obtained by calculating the similarity query image and keyframe image. The algorithm can automatically extract and describes the features and video content without human intervention. The algorithm can achieve video retrieval implementation independently.Video keyframe extraction and keyframes database building are used to test the retrieval performance of the algorithm. Experimental results verify that the algorithm can implement video search independently, and achieve accurate and objective retrieval. It has a high recall and precision.更多还原