基于SIFT特征点的视频镜头突变检测改进算法  

Improved algorithm for shot mutation detection based on SIFT feature points

作　　者：李姗姗 丰洪才[2] 苏筱涵 LI Shan-shan;FENG Hong-cai;SU Xiao-han(School of Mathematics and Computer Science ,Wuhan Polytechnic University,Wuhan 430023,China;School of Network Center,Wuhan Polytechnic University,Wuhan 430023,China)

机构地区：[1]武汉轻工大学数学与计算机学院,湖北武汉430023 [2]武汉轻工大学网络中心,湖北武汉430023

出　　处：《武汉轻工大学学报》2019年第1期67-72,共6页Journal of Wuhan Polytechnic University

摘　　要：该文提出了一种改进的结合SIFT特征点提取的视频镜头突变检测算法。针对SIFT算法特征描述子维数过高的问题,该文在SIFT算法基础上重新划分特征点邻域,将特征描述子维数降低50%。实验结果表明,改进的SIFT算法视频镜头突变检测平均查全率达到了97. 84%,查准率达到了96. 83%,比文献值分别高出2. 05%和2. 38%,平均每秒完成特征点提取的视频帧数为42. 4187,每一秒的特征点提取效率提高了60. 49%。提高了镜头变化检测的精度和时间效率。In the paper,an improved video shot mutation detection algorithm based on SIFT feature extraction is proposed.Aiming at the problem that the sub-dimension of SIFT algorithm is too high,this paper re-divides the feature point neighborhood based on SIFT algorithm and reduces the feature descriptor sub-dimension by 50%.The experimental results show that the average recall rate of the video shot mutation detection of the improved SIFT algorithm reaches 97.84%,and the precision rate reaches 96.83%,which is 2.05%and 2.38%higher than the literature value respectively.The number of video frames extracted by the point is 42.4187,and the feature point extraction efficiency per second is improved by 60.49%,improving the accuracy and time efficiency of lens change detection.

关 键 词：SIFT 特征点匹配 镜头突变检测 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]