奚晓晔,严利民,杜斌.帧间差值分布和渐变模型的视频镜头分割方法[J].电子测量与仪器学报,2016,30(11):1765-1773
帧间差值分布和渐变模型的视频镜头分割方法
Video shot segmentation based on the distribution of interframe difference and the gradient model
  
DOI:10.13382/j.jemi.2016.11.019
中文关键词:  镜头分割  帧间差值分布  渐变模型
英文关键词:video shot segmentation  the distribution of interframe difference  gradient model
基金项目:国家自然科学基金(61376028)资助项目
作者单位
奚晓晔 上海大学机电工程与自动化学院上海200072 
严利民 上海大学微电子研究与开发中心上海200072 
杜斌 上海大学微电子研究与开发中心上海200072 
AuthorInstitution
Xi Xiaoye School of Mechatronics Engineering and Automation, Shanghai University, Shanghai 200072, China 
Yan Limin Microelectronics R&D Center, Shanghai University, Shanghai 200072, China 
Du Bin Microelectronics R&D Center, Shanghai University, Shanghai 200072, China 
摘要点击次数: 82
全文下载次数: 132
中文摘要:
      视频镜头分割是视频挖掘和分析的重要一步,提出了一种帧间差值分布和渐变模型的视频镜头分割方法。先提出了一种减少帧间差运动量的方法,使帧间差更为理想化,通过对视频的帧间差序列求其值的分布得到检测部分突变帧的阈值,进而将整个帧间差序列进行分段,重复同样的步骤,最终得到所有的突变帧。在检测渐变帧方面,根据渐变过程中亮度的二阶差分的特点和渐变模型的特点,来求得正确的渐变帧。最后,在实验中,选取了不同类型的视频对本文的算法进行验证,最终证明了本文方法的正确性。算法的查全率和查准率都在80%以上,相比其他方法对于变化复杂的视频镜头检测查全率平均提高了5.74%,查准率平均提高了8.53%。
英文摘要:
      Video shot segmentation is an important step of video mining and analysis. This paper presents a method of video shot segmentation based on the distribution of interframe difference and the gradient model. At first, a method reducing the complexity of the interframe difference is proposed to make the interframe difference more ideal. This paper gets the threshold detecting a part of mutations frames through the interframe difference sequences, then piecewises the whole interframe difference sequence, and repeats the same steps to get all mutation frames finally. In the terms of detecting gradient frames, this paper uses the characteristics of the brightness second order difference of gradient process and the characteristics of gradient model to distinguish correct gradient frames. At last, this paper verifies the algorithm by different types of video in the experiment, and the result shows that the method proposed in this paper is correct. The recall and precision of the algorithm are more than 80%. Comparing with the other methods of video shot detection with complex change, the method proposed in this paper increases the average recall of 5.74% and increases the average precision of 8.53%.
查看全文  查看/发表评论  下载PDF阅读器