Video Copy Detection Based on Deep CNN Features and Graph-Based Sequence Matching

Xin Zhang,Yuxiang Xie,Xidao Luan,Jingmeng He,Lili Zhang,Lingda Wu

DOI: https://doi.org/10.1007/s11277-018-5450-x

IF: 2.017

2018-01-01

Wireless Personal Communications

Abstract:This paper introduces a novel content-based video copy detection method using the deep CNN features. An efficient deep CNN feature is employed to encode the image content while retaining the discrimination capability. Taking advantage of the extremely fast Euclidean distance similarity of deep CNN features, a keyframe-based copy retrieval method that exhaustively searches the copy candidates from the large keyframe database without indexing is proposed. Moreover, a graph-based sequence matching algorithm is employed to obtain the copy clips and accurately locate the video segments. The experimental evaluation has been performed to show the efficacy of the proposed deep CNN features. The promising results demonstrate the effectiveness of our proposed approach.