Abstract:

Key frame extraction is an important part of video summarization. The quality of key frame extraction directly affects people’s understanding of video. Most of the traditional key frame extraction algorithms are based on visual correlation, that is, only extracting the underlying information to calculate its similarity, ignoring the semantic correlation, which is easy to cause error and some redundancy. In this paper, a semantic-based video key frame extraction algorithm is proposed. Firstly, the algorithm uses hierarchical clustering algorithm to extract video key frames. Then, a semantic correlation algorithm is used to get rid of the redundant frames by histogram comparison. Finally, compared with other algorithms, the proposed algorithm has a relatively small redundancy.

Key words: hierarchical clustering algorithm, Hasche awareness, [k]-means, Scale Invariant Feature Transform（SIFT）, Bag of Feature（BOF）, semantic correlation

摘要：

视频关键帧提取是视频摘要的重要组成部分，关键帧提取的质量直接影响人们对视频的认识。传统的关键帧提取算法大多都是基于视觉相关的提取算法，即单纯提取底层信息计算其相似度，忽略语义相关性，容易引起误差，同时也造成了一定的冗余。对此提出了一种基于语义的视频关键帧提取算法。该算法首先使用层次聚类算法对视频关键帧进行初步提取；然后结合语义相关算法对初步提取的关键帧进行直方图对比，去掉冗余帧，确定视频的关键帧；最后与其他算法比较，所提算法提取的关键帧冗余度相对较小。

关键词: 层次聚类算法, 感知哈希, [k]-means, 尺度不变特征变换（SIFT）, BOF算法, 语义相关