基于核的监督非保局投影镜头边界检测

Shot Boundary Detection Based on a Kernel Supervised Non-Locality Preserving Projection

针对如何进一步提高镜头边界检测精度问题,提出了一种基于核的监督非保局(KSNLPP)投影视频镜头检测方法。在非保局投影(NLPP)中引入样本的类别信息,并利用核方法提升NLPP解决非线性问题的能力,提出了KSNLPP算法;在此基础上,将每个镜头视为一类,利用KSNLPP算法得到视频图像的投影矩阵,利用此投影矩阵对新的视频数据进行降维处理,实现有监督的视频特征提取;融合各帧的相邻帧特征构建中间特征,利用局部支持向量机(LSVM)分类器实现镜头边界检测。试验结果表明,提出的镜头边界检测方法能够显著提高镜头边界的检测精度。

Improving the precision of shot boundary detection is very important.This paper proposes a method on video boundary detection based on a kernel supervised non-locality preserving projection（KSNLPP）.Firstly,by introducing the class information into non-locality preserving projection（NLPP） and using the kernel method to solve the limitation problem of the nonlinear separability for NLPP,we get the kernel supervised non-locality preserving projection（KSNLPP）.Secondly,we gain the projection matrix of video feature by using KSNLPP where every sample shot is considered as a class,and the new video feature reduce to a lower one using the projection matrix.Finally,we build the intermediate features fusing the local temporal structure,and identify shot boundaries through inputting the intermediate features to two Local Support Vector Machine classifiers.The experimental results show that the proposed method can improve the precision of shot boundary detection.