基于二维EMD和非监督聚类的视频镜头边界检测

Video Shot Boundary Detection Algorithm Based on Two-dimensional EMD and Unsupervised Clustering Technologies

采用了改进的双三次和双线性混合插值和Delaunay三角剖分算法，分别得到极大值和极小值的包络面，通过包络面算法求出内蕴模态函数（IMF），以IMF作为图像的特征向量，进行非监督聚类分析。对特征向量进行分类，提取出样本中表示图像纹理特征的像素点的质心，从而确定镜头的边界。采用大量的镜头样本进行试验，结果表明，该方法有较高的查全率、查准率和鲁棒性。

The algorithms of improved bi-cubic and bilinear mixed interpolation and the Delaunay triangulation are adopted to get the enveloping surfaces of the maximum value and minimum value respectively. The Intrinsic Mode Function （IMF） can be obtained by means of the enveloping surface algorithm, which is taket~ as the image＇s eigenvector to implement the unsupervised clustering analysis. To classify the eigenveetors and extract the centroids of the pixels in the samples, which indicate the image texture＇s features, thus determine the boundary of the shot. A mass of shot samples are used for experiments, the test results show that the method is of higher recall ratio, precision ratio and robustness.