镜头检测是基于内容的视频检索和浏览首先要解决的关键技术。对视频结构和镜头变换进行概述,对典型的镜头检测算法进行研究分析,并提出了一种基于奇异值分解和特征融合的镜头检测算法。通过子空间投影得到帧的重构矩阵,计算帧与其重构矩阵的奇异值差,结合直方图特征进行镜头检测。实验结果表明了所提算法在镜头检测方面的有效性。
Shot detection is the key technology in the content-based video retrieval and browse. The video structure and the shot types are summarized. Several typical shot detection algorithms are analyzed and discussed. A shot detection algorithm based on SVD and feature-level fusion is proposed. The frame' s reconstructed matrix is given through the sub-space projection, and then the histogram feature as well as the SVD' s value difference between the frame and its reconstructed matrix is computed. A shot detection is processed after feature-level fusion. The experimental results prove the algorithm efficient.