摘要： 为解决网络视频的非法拷贝问题，提出一种基于峭度图像的视频指纹算法。对视频片段进行预处理后，利用均匀分布的随机变量提取关键帧以及关键帧的峭度图像，并对峭度图像进行离散余弦变换(DCT)，采用较大的DCT系数构造视频指纹，在视频指纹的匹配过程中，通过滑动窗的方法对不同长度的指纹进行匹配，从而达到视频认证的目的。实验结果证明，该算法提取的视频指纹在常见视频攻击下误码率均小于10%。

关键词: 基于内容的视频指纹, 视频哈希, 视频拷贝, 峭度图像, 离散余弦变换, 区分性

Abstract: In order to detect illegal copies of videos, this paper proposes a video fingerprinting algorithm based on kurtosis image. The video clip is preprocessed and key frames are selected by using a random variable with uniform distribution. The kurtosis images of key frames are computed and Discrete Cosine Transform(DCT) is applied into the kurtosis images. The video fingerprints are generated with the lager DCT coefficients. During video fingerprints matching, a sliding window method is used for matching, and the identification of videos is achieved. Experimental results prove that bit error rates are below 10% under common attacks.

Key words: Content-based Video Fingerprinting(CBVF), video hashing, video copy, kurtosis image, Discrete Cosine Transform (DCT), discrimination

中图分类号: 

• TP301.6