Video abstract selection method based on sparse kernel dictionary

• Main Authors: HE ROUBING, LIU AN'AN, SU YUTING, ZHANG JING
• Format: Patent
• Language: Chinese; English
• Published: 06.04.2018
• Subjects: CALCULATING; COMPUTING; COUNTING; HANDLING RECORD CARRIERS; PHYSICS; PRESENTATION OF DATA; RECOGNITION OF DATA; RECORD CARRIERS

The invention discloses a video abstract selection method based on a sparse kernel dictionary, and mainly overcomes extraction of video key frames. The algorithm comprises the following steps that step one, a video frame pool is formed by video extraction frames and multiple features of the video are extracted; step two, the Gaussian kernel matrix K and the residual error matrix P of the video areconstructed by using the features of the frames; step three, a reconstruction sparse sub-dictionary generating the video is optimized by using the SOMP (synchronous orthogonal matching iterative algorithm) and the constructed Gaussian kernel matrix; and step four, the obtained sparse representative sub-dictionary is utilized to be projected to the video frame pool and the key frames are selected. 本发明公开了种基于稀疏核字典的视频摘要选择方法，主要攻克视频关键帧的提取。算法包括以下步骤：第，将视频抽成帧构成视频帧池，提取视频的多个特征；第二，利用帧的特征，构建视频的高斯核矩阵K，和残差矩阵P；第三，利用SOMP(同步正交匹配迭代算法)和构建的高斯核矩阵优化生成视频的重构稀疏子字典；第四，利用得到的稀疏代表子字典投影到视频帧池，选择关键帧。