Fast and accurate content-based video copy detection using bag-of-global visual features

Abstract

In this paper, we propose a fast, accurate content-based video copy detection scheme based on bag-of-global visual features, which is characterized by (1) utilizing an efficient DCT-sign-based feature for fast detection; (2) performing multiple assignment in the temporal domain, in addition to the feature and spatial domain to ensure repeatability in segment-level matching; and (3) adopting an inverse document frequency weighting and temporal burstiness-aware scoring to emphasize distinctive visual words. Despite detection 95 times faster than real-time, the proposed system achieves a false negative rate of 0.2% against queries that are altered by non-geometric transformations without any false positives.

Recent advances in content based video copy detection

“…2D-DCT + BoVW [10] Global IDF weighting + Burstiness-aware scoring T3-T6,T8,T10 T1,T2,T7,T9 Time Accuracy Mean and variance of wavelet coefficients [11] Global Euclidean distance + Clustering based search S3,S5,S6, S11,S13,S14 S1,S9,S10,S15 Accuracy BPHOG + BGIST [14] Global Hamming distance + Copy confidence score T1,T2,T4-T10 T3 Accuracy MSF-color feature [15] Semi-global Edit distance based sequence matching S1-S5,S9 S11-S15 S6,S10 Time Spatial correlation descriptor [16] Global Chi-squared statistics + Edit distance S1-S11, S14,S15 S13 Accuracy [25] Temporal Matching using suffix array structure S1-S5,S9,S13-S15 S6,S10,S11 Time SIFT + Ordinal measure [27] Global+Local Transformation adaptive matching T1-T6,T8,T10 T1,T7,T9 Accuracy…”

“…Due to DCT, most of the energy will be converged in lower level frequencies, so this will reduce the total amount of data that is required to describe an image or video frame. Yusuke et al [10] perform feature extraction by applying 2D-DCT on each predefined block of keyframe to get AC coefficients, this DCT-sign based feature is used as signature of both reference and query video keyframes.…”

“…3) Yusuke et al [10] applied concept of bag-of-visualwords with DCT-sign based feature, which is usually used with local features. So applying concepts of local features with global ones can efficiently increase robustness of global features against various transformations.…”

Frame filtering and path verification for improving video copy detection

“…For example, the local descriptors are encoded by some projection method. Recently, some schemes like [3] [4] integrate utilize both the global and local feature in order to get balance between efficiency and effectiveness. For video matching, there are also two kinds of matching methods, i.e., matching based on the video sequence and matching based on frame fusion.…”