中文摘要：

图像中的Logo检测对于分析图像的内容、进行品牌广告投放和广告推荐具有重要的意义。针对现有的Logo检测方法存在的准确率低、处理速度慢的问题,提出了基于特征判别力分析和结构约束的Logo检测方法。首先,提出了基于出现频率的判别力分析方法;其次,提出了基于特征之间相对距离、相对主方向和相对尺度的结构关系表示方法,并构建出Logo表示模型;最后,提出了由粗到精的Logo检测方法,采用视觉单词判别力分析获得候选区域,并采用结构关系来进行精确匹配,确定最终的Logo区域。在一个包含100种Logo的10,000张图像的Logo数据集上的Logo检测实验中表明,所提出的方法在准确率、召回率和处理速度上均明显优于当前主流的Logo检测方法,证明了所提出方法的有效性和高效性。

英文摘要：

Logo detection in images is useful to image understanding, advertisement embedding and recommendation. Existing Logo detection methods suffer the problems of low accuracy and low processing speed. To address the problems, this paper proposes a visual word discrimination analysis and structure constraint based Logo detection method. First, this paper presents a frequency based discriminative power analysis method. Second, with analysis of the relative distance, relative principle orientation and relative scale of two local features, this paper proposes a structure based representation, and combines the discriminative visual words and structure constrain to build a Logo representation model. Finally, this paper presents a coarse -to -fine framework for efficiently detecting the logo region in images. A Logo detection experiments on a 10, 000 images （100 Logo types） dataset is conducted, and the proposed method outperforms the state -of- the -art methods in terms of precision, recall and processing speed, which demonstrates the effectiveness and efficiency of the proposed method.