| 摘 要: 随着网络视频数量呈指数级增长,视频版权等问题愈发严重。相关的科研领域也诞生了许多方法用于解决近似重复视频检索和部分重复视频检测等问题。以往的方法通常是先提取帧级特征再进行时间对齐,这会导致时序信息的丢失以及性能浪费。为此提出将镜头分割技术引入部分重复视频检测方法,保留镜头内的时序信息,并省去了时间对齐步骤。通过在 VCDB数据集上的测试,全面地分析了各项参数对方法性能的影响。与传统方法对比,所提方法在召回率上达到0.78,表现出色,提升约50%。该方法使用的镜头级相似度匹配能够减少性能需求并出色胜任部分重复视频检测任务。 |
| 关键词: 部分重复视频检测 镜头分割 混合卷积神经网络 |
|
中图分类号: TP37
文献标识码: A
|
|
| Combining Shot Segmentation to Detec tDuplicate Segmentsin Videos |
|
TAO Yuhan, CHEN Danwei
|
(School of Computer Science, Nanjing University of Post and Telecommunication, Nanjing 210042, China)
1072471078@qq.com; chendw@njupt.edu.cn
|
| Abstract: With the exponential growth of online videos’amount, issues such as video copyright infringement have become increasingly severe. In response, the research community has developed numerous methods to address problems like nea-r duplicate video retrieval and partial duplicate video detection. Traditional approaches often involve extracting frame-level features followed by temporal alignment, which can lead to the loss of temporal information and inefficiencies in performance. To address these limitations,this paper proposes the integration of shot segmentation technology into partial duplicate video detection, thereby preserving temporal information within shots and eliminating the need for temporal alignment. Through testing on the VCDB dataset, a comprehensive analysis of the impact of various parameters on the method’s performance was conducted. Compared to traditional methods, the proposed approach achieves a recall of 0.78, demonstrating a significant improvement of approximately 50% . The sho-t level
similarity matching employed in this method reduces performance requirements and effectively accomplishes partial duplicate video detection tasks. |
| Keywords: partial duplicate video detection shot segmentation hybrid convolutional neural network |
