基于改进BEMD的视频镜头转场检测算法研究  

Video shot detection algorithm based on improved BEMD

作　　者：戴小文[1] 魏志强[1] 苟先太[1] 

机构地区：[1]西南交通大学电气工程学院,四川成都610031

出　　处：《光电子．激光》2010年第2期270-273,共4页Journal of Optoelectronics·Laser

基　　金：信息产业部电子发展基金资助((2005).555)

摘　　要：对二维经验模式分解(BEMD)算法进行了改进,采用限定域的Delaunay三角剖分和三次插值得到极大值和极小值包络面,用基于限邻域经验模式分解(NLEMD),即通过设定最大邻域(时宽)和采用邻域内局部自适应均值算法代替包络均值算法进行分解,给出了图像BEMD分解后内蕴模函数(I MF)1和2的Hilbert谱,以I MF2的瞬时振幅作为图像的特征向量,计算镜头转场中图像序列帧特征向量间的欧式距离。采用大量的视频镜头转场的样本进行实验,结果表明,剪切镜头查准率和查全率皆为98%;渐变镜头查准率86.4%,查全率87.6%。A video shot boundary detection algorithm is presented by using improved bidimensional empirical mode decomposition（BEMD）.The image is extended to eliminate the border effect of BEMD and improve the BEMD.The domain limited Delaunay triangulation cubic interpolation is used to get the minimum envelope face and maximum envelope face.The BEMD is based on the neighborhood limited EMD（NLEMD）which dissolves the image by setting the max neighborhood,and using local adaptive means algorithm to replace the envelope means algorithm.The Hilbert spectra of IMF 1 and IMF 2 are given after BEMD.The instantaneous amplitude of IMF 2 is used as the characteristics of the frame.The Euclidean distance between characteristics of the sequence images is proposed.Experiment results indicate that the ratio of correct detection and detected ratio are all 98% for cut transitions,and 86.4% and 87.6% respectively for gradual transitions.

关 键 词：视频镜头 检测 二维经验模式分解(BEMD) HILBERT变换 

分 类 号：TN919.8[电子电信—通信与信息系统]