视频流中标版类广告的检测方法研究

发布时间：2019-05-20 14:29

【摘要】：随着商业化步伐的加快，现代媒体领域中的广告的元素日趋增多，媒体产业的发展也对视频监测工作提出了新的更高的要求。作为媒体市场监管方的视频监测部门常常需要检测视频中某个广告图标出现的时间位置和存在的时间长度，以此规范各广告商的经营行为。因此，相关方面的技术已是当今传媒领域的热门，也是媒体产业得以健康发展的必要保证。由于相关技术涉及图像处理、图像理解等新兴领域，缺乏现有方法和理论依据，因此对上述技术的研究也具有相当的难度。比如说，广告图标在视频中的显示常常伴随有尺寸、旋转、光照等变化，若使用依据像素的比较方法或基于内容的图像检测方法则无法适应如此多样的变化。另外，视频作为图像在时间轴上的序列，增加时间维度后也为检测工作增加了难度。方法和理论的匮乏阻碍了相关领域的研究，也阻碍了相关产业的发展。而本文以SURF方法为理论基础和出发点，先介绍SURF方法的步骤和基本原理，随后提出结合角点检测与SURF描述子的图像检测方法，并对其进行适应于标版广告图标和视频检测的改进，发展出一套检测视频中广告图标的方法，并将其用于视频检测；而后设计视频检测软件，并在实验中验证其有效性。据此，本文主要包括以下内容： 1、从图像间的匹配与比较的角度出发，并考虑到图像间可能存在的尺寸、旋转、视角、噪声变化等变化因素，选用对上述变化均有适应性和鲁棒性的SURF算法进行研究，介绍SURF算法基本步骤与原理。 2、出于快速性考虑，从SURF方法出发，提出结合角点检测与SURF描述子的图像检测方法。在简单介绍各类角点检测方法后，将角点检测与计算SURF描述子结合，形成新的图像检测方法，在保持快速性的前提下赋予角点尺度信息，解决图像匹配的尺度变化适应性问题，使用实际广告图标例子对图像检测方法进行实例验证，并与传统SURF方法比较，分析其优越性。 3、由于图像检测方法在标版类广告图标的匹配工作中仍不甚理想，本文根据广告图标的特殊性对图像检测方法进行匹配比计算和多模板图像匹配算法等多方面改进，以更适用于对视频中标版类广告图标的检测工作。 4、在对方法进行适应于广告图标的若干改进后，需将方法应用于实际的视频检测工作中。本文在此先介绍视频的结构，并在此基础上针对视频的特殊性对方法加以改进，提出了算法用于视频的具体方法，以达到检测视频中标版类广告图标的目的，并分析方法的有效性。 5、最后依照软件工程的相关理论，使用基于数据流的方法设计视频检测软件。并通过软件操作的演示，使用实例验证，，以最终达到准确、实时地检测视频中广告图标的目的。
[Abstract]:With the acceleration of commercialization, the elements of advertising in the field of modern media are increasing day by day, and the development of media industry has also put forward new and higher requirements for video monitoring. As the regulator of the media market, the video monitoring department often needs to detect the location and length of the appearance of an advertising icon in the video, in order to standardize the business behavior of each advertisers. Therefore, the related technology has become a hot topic in the field of media, and it is also a necessary guarantee for the healthy development of the media industry. Because the related technologies involve image processing, image understanding and other emerging fields, lack of existing methods and theoretical basis, so the research on the above technologies is also quite difficult. For example, the display of advertising icons in video is often accompanied by changes in size, rotation, lighting and so on. If the pixel-based comparison method or content-based image detection method is used, it can not adapt to such a variety of changes. In addition, as a sequence of images on the timeline, video increases the difficulty of detection after increasing the time dimension. The lack of methods and theories not only hinders the research in related fields, but also hinders the development of related industries. Based on the theory and starting point of SURF method, this paper first introduces the steps and basic principles of SURF method, and then proposes an image detection method combining corner detection and SURF descriptive. It is suitable for the improvement of standard advertisement icon and video detection, and develops a set of methods to detect the advertisement icon in video, and applies it to video detection. Then the video detection software is designed and verified in the experiment. According to this, this paper mainly includes the following contents: 1, from the point of view of image matching and comparison, and taking into account the possible size, rotation, angle of view, noise change and other factors between images, The SURF algorithm, which has adaptability and robustness, is selected to study the above changes, and the basic steps and principles of SURF algorithm are introduced. 2. For the sake of rapidity, an image detection method combining corner detection and SURF descriptive is proposed based on SURF method. After briefly introducing all kinds of corner detection methods, a new image detection method is formed by combining corner detection with computational SURF description. Under the premise of maintaining rapidity, corner scale information is given to solve the adaptability of image matching to scale change. The actual advertising icon example is used to verify the image detection method, and compared with the traditional SURF method, its advantages are analyzed. 3. Because the image detection method is still not ideal in the matching of typesetting advertising icons, this paper improves the image detection methods in many aspects, such as matching ratio calculation and multi-template image matching algorithm according to the particularity of advertising icons. It is more suitable for the detection of video winning advertising icons. 4. After some improvements of the method to adapt to the advertisement icon, the method should be applied to the actual video detection work. In this paper, the structure of video is introduced, and on this basis, the method is improved according to the particularity of video, and the concrete method of using algorithm for video is put forward in order to detect the advertisement icon of the winning version of video. The effectiveness of the method is analyzed. Finally, according to the related theories of software engineering, the video detection software is designed by using the method based on data flow. Through the demonstration of software operation, an example is used to verify that the advertising icon in video can be detected accurately and in real time.
【学位授予单位】：上海交通大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.41

【参考文献】