文献类型：期刊文献

中文题名：基于渐进二分策略的自适应阈值视频镜头检测

英文题名：Shot boundary detection algorithm of adaptive threshold based on progressively bipartite strategy

作者：霍奕[1,2];王艳峰[2];杨楚翘[1]

第一作者：霍奕

机构：[1]北京联合大学师范学院;[2]中国传媒大学新媒体研究院

第一机构：北京联合大学师范学院

年份：2018

卷号：38

期号：A01

起止页码：198-201

中文期刊名：计算机应用

外文期刊名：Journal of Computer Applications

收录：CSTPCD;;北大核心:【北大核心2017】；CSCD:【CSCD_E2017_2018】；

语种：中文

中文关键词：视频镜头边界检测;渐变检测;渐进二分策略;自适应阈值;神经网络模型

外文关键词：shot boundary detection;gradual transition detection;progressively bipartite strategy;adaptive threshold;neural network model

摘要：针对目前视频镜头边界检测算法多专注某特定类型,没有一种对于各类渐变类型检测都适用的算法的问题,以及在检测时间方面,现有的视频镜头检测算法需要对所有帧进行计算,因而具有很高时间复杂度的问题,提出一种新颖的基于渐进二分策略和自适应阈值的视频镜头分割算法。首先,它根据每个视频自身的特征用神经网络模型训练阈值计算参数,以自适应地生成检测阈值来提高检测准确性;然后,它采用渐进地二分策略进行视频帧间差计算,实现对各类渐变类型检测都使用统一的方法,并同时降低了计算时间复杂度。实验表明本算法在检测性能上,对切变检测的准确率提高了5. 36%,对渐变检测的准确率提高了9. 13%;在计算复杂度上降低了27. 32%。本算法在检测性能和计算复杂度上均具有显著的优越性,并对各类渐变类型的镜头检测都适用。
Most of the existing gradual transition detection algorithms focus on some special types, and there is not any algorithm suitable for all types of gradual transitions. Current shot boundary detection algorithms also have high computation complexity since they have to perform in sequence. For these problems, this paper proposes a novel shot boundary detection algorithm with adaptive threshold based on progressively bipartite strategy. Firstly, It adaptively produced thresholds according to the characteristics extracted from each video to improve detection quality. Secondly, it used progressively dimidiate strategy to compute in order to detect various types of gradual transitions by an unified approach and reduce computation complexity at the same time. Experimental results show that it improves the accuracy of cut transition detection by 5.36%, improves the accuracy of gradual transition detection by 9.13%, and reduces computational complexity by 27.32%. The algorithm is superior in both detection quality and computation complexity, and is suitable for all types of gradual transition detection.