【摘要】 随着视频媒体中的广告元素日趋增多,对角标广告的出现时间和存在时长的检测成为视频监播工作新的需求。角标广告在视频中呈现形式多样,出现时间具有随机性,并且视频相邻帧之间的图像内容可能存在较大差别。针对这一特点,本文以主方向模板特征为基础提出视频中角标广告实时检测算法。该方法建立特定的匹配模板剔除目标图像中背景部分的干扰,并引入加权模板抑制角标边缘可能造成的误匹配,同时主方向模板对图像的特征压缩可以克服传统模板匹配计算复杂度高的缺点。实验结果表明,本方法同时兼顾了检索性能与实时处理能力,达到了视频中角标广告的检测要求。更多还原

【Abstract】 With the growing number of video media advertising elements, the detection of time and length of Advertisement Logo has become a new demand. Advertisement logos vary in shape features with their occurrence of randomness,and image contents between adjacent video frames may exist obvious difference. Considering such characteristics,based on dominant orientation templates( DOT),this paper ventures to put forward the real-time detection algorithm of Advertisement Logo in videos. The algorithm establishes specific matching template to eliminate the background interference in the target image,and introduce the weights template to suppress the generation of mismatching caused by the edge of the logo,at the same time DOT’s feature compression on images can overcome the defect of high computation complexity shared by the traditional template matching method. The experimental results prove that the presented approach can balance retrieval performance with efficiency,and meets the requirements of Advertisement Logo detection in video.更多还原