结合角点和局部特征描述的标版广告检测算法  

Advertisement Logo Detecting Approach Combining Corner and Local Feature Description

作　　者：徐聪[1] 李炎[1] 杜秀华[1] 曹俊 

机构地区：[1]上海交通大学自动化系系统控制与信息处理教育部重点实验室,上海200240 [2]上海东方娱乐传媒集团广告经营中心,上海200041

出　　处：《实验室研究与探索》2014年第4期108-111,194,共5页Research and Exploration In Laboratory

摘　　要：针对媒体中广告元素检测需求的多样化,通常需要检测视频或图像中是否存在相应的标版类广告,提出了一种结合Harris角点和SURF描述子的广告图标检测算法。算法利用了Harris角点物理意义明确、稳定性、鲁棒性好、提取用时较快等优点,用于提取图像的特征点;并利用SURF描述子具有的良好的特征描述作用,以及对旋转、亮度等变换的适应性,对特征点生成多个不同尺度的SURF描述子,以用于特征描述和图像间的比较,从而解决图像匹配中的尺度变化适应性问题。使用实际用例对方法进行验证,结果表明方法较之于SURF算法具有更好的快速性,且效果良好,检测结果准确。Because of the diverse requirement of detecting advertisement elements in media, advertisement logos always need to be detected in videos or images. So, this paper gives an advertisement logo detecting approach combining Harris corner and SURF descriptors. This approach takes advantage of clear physical meanings, good stability and robustness, fast extraction of Harris corner, to extract the feature points of an image; and takes advantage of good characteristic of SURF descriptor in feature description and adaption to rotation, lightness and other variations, to generate some SURF descriptors in different scales for a feature point, and applies them to feature description and comparison between images, to eventually solve the problem of scale-variance adaptability in image matching. By testifying the approach with examples, the result shows that the approach is of better rapidness than SURF, the performance is accurate and good

关 键 词：标版广告检测 HARRIS角点 SURF方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]