基于图像识别的电视广告监播系统

【作者】 贾军；

【导师】 张重雄；

【作者基本信息】 南京理工大学 ， 通信与信息系统， 2008， 硕士

【摘要】 随着经济的发展,电视广告成为社会生活中越来越重要的一部分,而其带来的社会问题也日渐显著,特别是虚假广告严重误导了消费者,坑害了广大人民,因此广告监测成为社会急需处理的问题。本设计介绍了一种基于图像识别的电视广告监播系统。该系统结合图像采集、二值化预处理等技术,以图像模式识别的方式实现对电视广告的监播。本设计主要研究广告图像识别技术的软件实现。系统使用视频采集卡,采用DirectShow技术对电视信号进行采集并捕捉特定的帧用于后期识别。针对视频广告识别的特点和要求,采用了“首尾重点提取法”关键帧提取方法。在采集到电视广告关键帧后,为了简化运算,需要对选取的关键帧进行二值化预处理。系统的二值化模块中,采用了对不同亮度情况的图像处理效果都较好,而且算法比较稳定的Otsu算法。由于需要识别的对象,也就是需要监播的广告是事先已知的,因此可以事先提取样本,采用基于模板的匹配方法进行识别,本设计选择了SSDA算法。由于在识别模块中,待识别的对象都是二值图像,本设计对SSDA算法进行了适当的调整,使其在识别二值图像时效率更高,并且加入了绝对误差值最低累加次数这一限制,以保证图像识别的准确性。在上述工作的基础上,本课题在VC环境下进行了对指定广告图像的测试实验。通过实验表明,该方法对广告监测有着一定的实用意义。更多还原

【Abstract】 TV Advertising is an important part of daily life, by which people know what products they like or dislike. But as the important effects it takes on, it is more and more urgent for us to monitor whether the right ads is on TV. This design introduces an automatic picture oriented television advertisement monitoring system. The system implements television advertisements monitoring in the way of pattern recognition together with image capturing and image binarization technologies.In this paper, the real-time realization of picture oriented Recognition technology on common platform is investigated. The system captures television signal with video capture card by DirectShow technique. Aiming at the characteristic and the request of the television advertisements, the system introduces the way of capturing the key-frames of the beginning and the end mainly.After the original data collection step, binarizing the captured images is needed to make the following steps more efficient. In the binarization module, the system introduces the Otsu algorithm, which is better effective to different brightness of the image and is steadier.Since the monitored objects, that is, the advertisements, are known, the system can extract samples of the advertisements, and use algorithms based on template match to recognize these advertisements. SSDA algorithm is adopted in this system. Because the pictures are all binary images, this article adjusts SSDA algorithm so that it can process binary images more efficiently, and an additional limitation, the absolute error accumulation times, are affiliated to insure the recognition correctness.Based on the working above, a set of picture oriented recognition system and the corresponding demo software are built under VC. The system is examined with some named advertising. The results show that the system has a higher recognition rate.更多还原