基于相似性保持和判别性分析的快速视频哈希算法  

A fast video hashing algorithm based on similarity preservation and discrimination analysis

作　　者：卞小曼 王小琴 蓝如师 刘振丙 罗笑南 BIAN Xiaoman;WANG Xiaoqin;LAN Rushi;LIU Zhenbing;LUO Xiaonan(School of Computer Science and Information Security,Guilin University of Electronic Technology,Guilin 541004,Guangxi,China)

机构地区：[1]桂林电子科技大学计算机与信息安全学院,广西桂林541004

出　　处：《山东大学学报（工学版）》2023年第6期63-69,81,共8页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(62172120,61772149,61936002,6202780103);广西科技计划资助项目(2019GXNSFFA245014,AD18281079,AA18118039,AD18216004);广西图像图形与智能处理重点实验室开发课题(GIIP2001)。

摘　　要：针对当前视频哈希算法检索时间长、准确率低的问题,提出一种基于相似性保持和判别性分析的快速视频哈希算法。通过视频的标签信息直接得到相似性矩阵,减少视频关键帧之间相似度的计算时间。采用根据标签信息定义的码本矩阵,使同一类别的视频生成相同的哈希码。在算法优化过程中,使用迭代优化找到参数的封闭解,学习得到视频哈希函数。本研究使得相似性矩阵和码本矩阵共同作用,不仅获得了较高的准确率,还极大提高了检索效率。试验结果表明,在两个公开视频数据集HMDB51和UCF101上,在数据集设置相同的情况下,该算法和6个常用的哈希算法相比,检索的时间和空间复杂度都明显优于其他算法。To address the problem of long retrieval time and low accuracy of current video hashing algorithms,a method named fast video hashing algorithm based on similarity preservation and discrimination analysis was proposed.The similarity matrix was directly obtained by the label information of the video dataset to reduce the calculation time of the similarity between the video keyframes.A codebook matrix defined based on tag information enabled videos of the same category to generate the same hash code.In the process of algorithm optimization,iterative optimization was used to find the closed solution of the parameter and learn to obtain the video hash function.The similarity matrix and codebook matrix worked together,which not only obtained high accuracy,but also greatly improved the retrieval efficiency.The experimental results on two public video datasets,e.g.,HMDB51 and UCF101,demonstrated that the time and spatial complexity of retrieval of this algorithm were significantly better than other algorithms compared with the six commonly used hashing algorithms when the dataset settings were the same.

关 键 词：快速学习 相似性保持 判别性分析 视频检索 机器学习 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]