【摘要】 针对基于内容的视频检索领域中,关键帧特征矩阵维度不同时的相似度计算问题,提出一种基于主成分分析的关键帧相似度核聚类检索算法。首先,针对任意具有不同数量关键帧的视频片段,提取特征向量并构造不同维度的特征矩阵。其次,基于PCA计算对特征矩阵进行SVD计算降维矩阵后,结合矩阵运算方法及核方法设计出一种视频关键帧相似度核聚类检索算法,并给出其加权改进形式。最后,通过测试视频标准库和人工视频片段的实验表明,该算法能更好地提视频高视频检索的效率。更多还原

【Abstract】 In the content-based video retrieval research, a PCA based key frame similarity kernel clustering algorithm is proposed to calculate the similarity of the feature matrix of video key frame with different dimensions. Firstly, feature vectors and structure feature matrices with different dimensions of any different video clip key frame are extracted. Secondly, the dimension reduction matrix with SVD method based on PCA algorithm is calculated, the key frame similarity kernel clustering algorithm is proposed with the matrix calculation method and the kernel method, and its improved weighted representation is proposed as well. Finally, the simulation experiments on the standard test video database and artificial video clip database show that the algorithm can improve the efficiency of video retrieval.更多还原