一种新的镜头边界检测和静态视频摘要提取方法

Novel approach of shot boundary detection and static video abstract extraction for video flow

提出一种新的融合图像分割和注意力模型的静态视频摘要自动生成方法.该方法将视频中的每帧图像分割成块,分别计算连续2帧对应块的匹配距离;在此基础之上,根据各个块不同的重要性用线性融合的方法求出2帧之间的匹配距离;然后通过计算匹配距离的期望和标准差自动地求出阈值,进而检测出镜头边缘;最后根据每个镜头的匹配距离,提取出静态视频摘要.仿真结果表明:与同类方法相比,该方法能够快速高效地检测出各种镜头边界,提取的静态视频摘要能够有效反映视频的内容.

A novel automatic approach of generating static video abstract is proposed by combining image segmentation with attention model. A frame is firstly divided into many blocks, then the matching difference of corresponding blocks between two consecutive frames is computed. The difference of two frames is computed with linear fusion scheme according to the different importance of all blocks. After the computation of expectation and standard deviation, threshold is worked out automatically and shot boundaries are detected. Finally, the matching difference of every shot is worked out, and representative frames are extracted. The simulation results show that the proposed method outperforms other methods in efficiency and effectiveness. The representative frames can represent the content of video sequences.