基于 UniformLBP 特征的镜头检测算法

摘要

首次将 UniformLBP 特征引入镜头检测中，提出一种有效的镜头检测算法。算法的主要思想是：以 UniformLBP 特征作为描述视觉信息的主要特征，利用图模型计算图像序列的特征差异信号，采用主动处理策略构造特征向量和选择训练数据，通过 SVM分类器识别视频镜头。提出的算法以 TrecVid05实验数据为测试平台，通过与 TrecVid05的测试结果比较可以证明该算法不但对切变镜头检测有很好的实验结果，而且能更为有效地检测渐变镜头。%In this article,we propose an effective shot detection algorithm,and it is the first time to introduce UniformLBP feature to shot detection.The main ideas of the algorithm are:First,it uses UniformLBP feature as the principal feather to describe visual information. Secondly,it employs graph-model to calculate the feature difference signal of image sequences.Thirdly,it adopts active learning strategy to construct eigenvector and choose training data.Finally,it identifies the video shot boundary by SVMclassifier.For the algorithm proposed in the paper,we use TrecVid05 experimental dataset as the test platform,and through comparison it with the test result of TrecVid05,it can prove that this algorithm not only has good experimental result on cut shot boundary detection,but can also more effectively detect the gradual shot transition case.