【摘要】 针对已有的视频拷贝检测算法仅使用单一特征进行视频内容匹配,难以应对多种不同形式的拷贝变化的问题,提出一种基于多种视觉特征的视频拷贝检测算法.该算法采用级联式检测过滤框架,在提取视频帧图像的全局特征用于检测画面轻微变化的重复视频片段后,使用更精准的局部特征等来检测各种复杂变化后的拷贝内容.为在大规模数据库中实现快速检测,使用kd树型索引结构实现特征近邻检索.在标准评测数据集上的实验结果表明,文中算法对多种拷贝变化具有鲁棒性,并具有较高的检测效率.更多还原

【Abstract】 Video copy detection is an important technique for filtering out unnecessary video content which is copyright invaded or redundant for large video databases.The difficulties of detecting video copies lie in the dissimilarities between video and its copies.Various visual transformations make the existing methods invalid where a single feature is used for video matching.This paper proposes a novel method which uses multiple visual features for content matching and searching.The proposed method adopts the cascade detection approach which first retrieves the query video using global features,then videos is matched by more accurate local features.To achieve satisfactory efficiency for large scale applications,the kd-tree indexing structure is adopted.Experiments on benchmark dataset show that the proposed method manifests a good balance of robustness to common copy transformations and high detection efficiency.更多还原