Perceptual Robust Hashing for Video Copy Detection with Unsupervised Learning
Open accessAbstract
In this paper, we propose an end-to-end perceptual robust hashing scheme for video copy detection based on unsupervised learning. Firstly, the spatio-temporal information in videos is effectively fused and condensed into high-dimensional features through a 3D self-attention, multi-scale feature fusion model based on 3D-CNN, in which the Inception block and the 3D self-attention mechanism are integrated. Then, we calculate the correlation distances between the extracted features to differentiate perceptual contents. Based on the similarity relationship, we can dynamically generate the pseudo-labels and exploit them to further guide the model training for video hash generation. In addition, we design the dual constraints to make the hash code obtain satisfactory robustness and discrimination. Extensive experiments demonstrate that the proposed scheme achieves superior performance of copy detection compared with existing schemes and performs well even in the case of untrained manipulations.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
2 of 13 cited sources checked · 0 supported · 0 flagged · accuracy not yet measured
- Premisecitation
Deep learning techniques can effectively extract the features of videos as long as a proper model and loss function are designed.
“Numerous studies [14,15] have shown that deep learning techniques can effectively extract the features of videos as long as a proper model and loss function are designed.”
Quote in the paperSubstantial claim1 of 2 sources checked- cites [14] 14. X. Luo, H. Wang, D. Wu, et al.. "A survey on deep hashing methods." ACM Transactions on Knowledge Discovery from Data (TKDD), 2022, pp. 1556-4681. source
The Introduction and §3.1 say that deep supervised hashing depends on network architecture and loss-function design, and §5.3 includes video among the modalities addressed by multi-modal deep hashing. The paper does not establish the more specific claim that a suitable model and loss function are sufficient to extract video features effectively; its discussion concerns learning hash codes for retrieval. The indexed passages provide section labels but no page numbers. (LLM, conf=0.81)
Experimental — accuracy not yet measured - cites [15] 15. C. Qin, E. Liu, G. Feng, and X. Zhang. "Perceptual image hashing for content authentication based on convolutional neural network with multiple constraints. " IEEE Transactions on Circuits and Systems for Video Technology, vol. 31, no. 11, pp. 4523-4537, 2021. source
- Findingown data
For most manipulations, the F1 scores of the proposed scheme are greater than those of the other compared schemes.
“We can find that, for most manipulations, the 𝐹 1 scores of our scheme are greater than those of other compared schemes.”
Quote in the paperSubstantial claim0 of 3 sources checked- cites [23] 23. R. Anuranji and H. Srimathi. "A supervised deep convolutional based bidirectional long short term memory video hashing for large scale video retrieval applications. " Digital Signal Processing, vol. 102, pp. 102729, 2020. source
- cites [24] 24. K. Wang, C. Cheng, Y. Chen, Y. Song, and S. Lai. "Attention-based deep metric learning for near-duplicate video retrieval. " In 2020 25th International Conference on Pattern Recognition (ICPR), 2021, pp. 5360-5367. source
- cites [25] 25. Y. Hu and X. Lu. "Learning spatial-temporal features for video copy detection by the combination of CNN and RNN. " Journal of Visual Communication and Image Representation, vol. 55, pp. 21-29, 2018. source
- Findingown data
When eta is set to 2, the performance of the proposed video hashing reaches the best.
“when 𝜂 is set as 2, the performance of our video hashing reaches the best at this situation.”
Quote in the paperNot a substantial claim - Findingown data
The hash distance of distinct videos is significantly larger than that of identical videos.
“Compared with the statistics of robustness results in Table 1, we can see that the hash distance of distinct videos is significantly larger than that of identical videos, thus proving the discrimination performance of our scheme.”
Quote in the paperSubstantial claim - Findingown data
The proposed scheme performs well even in the case of untrained manipulations.
“Extensive experiments demonstrate that the proposed scheme achieves superior performance of copy detection compared with existing schemes and performs well even in the case of untrained manipulations.”
- Also stated in: § 4.3 Performance Comparison · hedged
“It should be noted that the performance of our scheme does not degrade much in the case of untrained manipulations, which indicates the strong generalization ability.”
Quote in the paperSubstantial claim - Also stated in: § 4.3 Performance Comparison · hedged
- Findingnone given
The proposed scheme adapts to more video content-preserving manipulations without labeling and training.
“Our scheme can not only achieve satisfactory robustness and discrimination, but also adapts to more video contentpreserving manipulations without labeling and training.”
Quote in the paperNot a substantial claim - Findingown data
When alpha is set to 0.5, the L2 loss brings an obvious improvement that is better than other parameter values.
“We can find that, when 𝛼 is set as 0.5, the L 2 loss can bring a obvious improvement that is better than other parameter values”
Quote in the paperNot a substantial claim - Findingown data
The proposed scheme achieves superior copy detection performance compared with existing schemes.
“Extensive experiments demonstrate that the proposed scheme achieves superior performance of copy detection compared with existing schemes and performs well even in the case of untrained manipulations.”
- Also stated in: § 4.3 Performance Comparison · boosted
“As shown in Fig. 5, the ROC curve of our scheme is the closest to the upper left corner, which means our scheme achieves the best overall performance with respect to perceptual robustness and discrimination.”
Quote in the paperSubstantial claim0 of 3 sources checked- cites [23] 23. R. Anuranji and H. Srimathi. "A supervised deep convolutional based bidirectional long short term memory video hashing for large scale video retrieval applications. " Digital Signal Processing, vol. 102, pp. 102729, 2020. source
- cites [24] 24. K. Wang, C. Cheng, Y. Chen, Y. Song, and S. Lai. "Attention-based deep metric learning for near-duplicate video retrieval. " In 2020 25th International Conference on Pattern Recognition (ICPR), 2021, pp. 5360-5367. source
- cites [25] 25. Y. Hu and X. Lu. "Learning spatial-temporal features for video copy detection by the combination of CNN and RNN. " Journal of Visual Communication and Image Representation, vol. 55, pp. 21-29, 2018. source
- Also stated in: § 4.3 Performance Comparison · boosted
- Findingown data
The proposed scheme achieves the best overall performance with respect to perceptual robustness and discrimination among the compared schemes.
“As shown in Fig. 5, the ROC curve of our scheme is the closest to the upper left corner, which means our scheme achieves the best overall performance with respect to perceptual robustness and discrimination.”
Quote in the paperSubstantial claim0 of 3 sources checked- cites [23] 23. R. Anuranji and H. Srimathi. "A supervised deep convolutional based bidirectional long short term memory video hashing for large scale video retrieval applications. " Digital Signal Processing, vol. 102, pp. 102729, 2020. source
- cites [24] 24. K. Wang, C. Cheng, Y. Chen, Y. Song, and S. Lai. "Attention-based deep metric learning for near-duplicate video retrieval. " In 2020 25th International Conference on Pattern Recognition (ICPR), 2021, pp. 5360-5367. source
- cites [25] 25. Y. Hu and X. Lu. "Learning spatial-temporal features for video copy detection by the combination of CNN and RNN. " Journal of Visual Communication and Image Representation, vol. 55, pp. 21-29, 2018. source
- Interpretationown data
The proposed scheme has strong generalization ability.
“It should be noted that the performance of our scheme does not degrade much in the case of untrained manipulations, which indicates the strong generalization ability.”
Quote in the paperSubstantial claim - Interpretationnone given
The proposed end-to-end unsupervised perceptual video hashing scheme can be used for content authentication and copy detection.
“An end-to-end unsupervised perceptual video hashing scheme is proposed that can be used for content authentication and copy detection.”
Quote in the paperNot a substantial claim - Interpretationnone given
The designed dual constraints make the hash code obtain satisfactory robustness and discrimination.
“In addition, we design the dual constraints to make the hash code obtain satisfactory robustness and discrimination.”
Quote in the paperNot a substantial claim - Interpretationreasoning
The proposed scheme is an end-to-end deep learning scheme that integrates feature extraction and hash mapping to generate perceptual spatio-temporal feature of the video.
“we can build an end-to-end scheme based on deep learning that integrates feature extraction and hash mapping to generate perceptual spatio-temporal feature of the video.”
Quote in the paperSubstantial claim - Interpretationnone given
The proposed perceptual robust hashing scheme for video copy detection is based on unsupervised learning.
“In this paper, we propose an end-to-end perceptual robust hashing scheme for video copy detection based on unsupervised learning.”
Quote in the paperSubstantial claim - Interpretationreasoning
The proposed scheme is more suitable for increasingly massive big data.
“Since our scheme does not require additional supervised information, it not only has superior copy detection performance compared with some state-of-the-art schemes, but also is more suitable for the increasingly massive big data.”
Quote in the paperNot a substantial claim - Interpretationnone given
The proposed scheme does not require additional supervised information.
“Since our scheme does not require additional supervised information, it not only has superior copy detection performance compared with some state-of-the-art schemes, but also is more suitable for the increasingly massive big data.”
Quote in the paperSubstantial claim
- May combine two claims. 9
- Classified as probably not a substantial claim. 3
May combine two claims.
PremisecitationCurrent perceptual hashing schemes often require manual annotation of video samples and their similarity relationships, and these methods may not have enough generalization capability to deal with massive video and emerging video manipulations.
“But, current perceptual hashing schemes often require manual annotation of video samples and their similarity relationships, and these methods may not have enough generalization capability to deal with massive video and emerging video manipulations [16].”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [16] 16. N. Behrmann, J. Gall, and M. Noroozi. "Unsupervised video representation learn- ing by bidirectional feature prediction." In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2021, pp. 1670-1679. source
Split into 2 atoms — 0 confirmed, 0 partially supported, 0 disputed, 0 misattributed, 2 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured
May combine two claims.
Findingnone givenThe proposed scheme has satisfactory robustness and discrimination.
“Overall, our scheme has satisfactory robustness and discrimination, and can be effectively used for video copy detection.”
- Also stated in: § Hash Code · plain
“Our scheme can not only achieve satisfactory robustness and discrimination, but also adapts to more video contentpreserving manipulations without labeling and training.”
Quote in the paperSubstantial claim- Also stated in: § Hash Code · plain
May combine two claims.
Findingown dataThe hash distances between perceptual distinct videos have a mean of 11.29 and a standard deviation of 5.37.
“From the statistical results of hash distances between perceptual distinct videos, we can find that the mean value is 11.29 and the standard deviation is 5.37.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataThe ROC curve of 32-bit hash length is the worst, and the ROC curve of 128 bits is the best, but it is not much different from that of 64 bits.
“It can be seen from Fig. 6(b), the ROC curve of 32 bits is the worst, and the ROC curve of 128 bits is the best, but it is not much different from that of 64 bits.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown dataThe two strategies, the 3D self-attention multi-scale feature fusion model and pseudo-label generation, both lead to a certain performance gain.
“From Fig. 6(a), it can be found that the two strategies both lead to a certain gain.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S4 (based). May combine two claims.
Interpretationnone givenThe designed 3D self-attention multi-scale feature fusion model can extract compact perceptual feature and highlight the important parts.
“we design a 3D selfattention multi-scale feature fusion model, which can extract compact perceptual feature and highlight the important parts.”
Quote in the paperNot a substantial claimMay combine two claims.
PremisecitationThe SHMB scheme is supervised and its hash length is 96 bits.
“Although the effect of SHMB [23] is approximate to ours, it is worth noting that SHMB scheme is supervised, and its hash length is 96 bits.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [23] 23. R. Anuranji and H. Srimathi. "A supervised deep convolutional based bidirectional long short term memory video hashing for large scale video retrieval applications. " Digital Signal Processing, vol. 102, pp. 102729, 2020. source
Probably not a substantial claim: may not pass S4 (based). May combine two claims.
Premisenone givenThe above schemes divided feature extraction and hash mapping into two parts, and both were designed artificially and empirically, which may not achieve the global optimal result.
“The above schemes divided feature extraction and hash mapping into two parts, and both were designed artificially and empirically, which may not achieve the global optimal result.”
Quote in the paperNot a substantial claimMay combine two claims.
InterpretationreasoningThe proposed scheme greatly saves labor cost, and its hash code has low redundancy, saving storage and computation cost.
“That is to say, our video hashing scheme not only has excellent performance for copy detection, but also greatly saves labor cost, while the hash code has low redundancy, saving storage and computation cost.”
Quote in the paperSubstantial claimProbably not a substantial claim: may not pass S4 (based).
Interpretationnone givenThe model training can be better guided and constrained due to the employment of pseudo-labels.
“In addition, the model training can be better guided and constrained due to the employment of pseudo-labels.”
Quote in the paperNot a substantial claim
The paper relies on this without support.
The correlation distance between video-level features reflects perceptual similarity, such that a larger absolute correlation coefficient means higher perceptual similarity of the two videos.
“𝜌 ∈ [-1, 1] is the correlation coefficient, and the larger the absolute value of 𝜌, the higher perceptual similarity of the two videos.”
The paper relies on this without support.
The pseudo-label matrix generated from correlation distances can serve as ground truth for similarity judgment.
“We exploit the obtained pseudo-label matrix S as the ground truth for similarity judgment.”
The paper relies on this without support.
The training set has a scarcity of perceptual identical videos.
“Due to the scarcity of perceptual identical videos in the training set, there is an extremely rare possibility that 𝑡 𝑟 ⩽ 𝑡 𝑙 or 𝑡 𝑙 ⩽ 0, in which case this batch does not participate in the training.”
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
Each contradiction between two passages listed here was confirmed by an independent check.
Stated correlation-distance range excludes negative correlations
MajorThe paper permits correlations as low as −1, but its definition d = 1 − ρ then permits distances as high as 2. The stated [0, 1] range is therefore inconsistent with the definition.
“𝑑 ∈ [0, 1] is the correlation distance.”
“𝑑 (Z 𝑖 , Z 𝑗 ) = 1 -𝜌 (Z 𝑖 , Z 𝑗 )(5) where Z 𝑖 and Z 𝑗 (𝑖, 𝑗 = 1, 2, ..., 𝑛) are the two video-level features of two videos 𝑣 𝑖 and 𝑣 𝑗 , respectively, and 𝑛 denotes the quantity of videos in the batch.𝜌 ∈ [-1, 1]”
Negative correlations receive the opposite similarity treatment
MinorThe paper says a larger absolute correlation means greater similarity. Yet d = 1 − ρ makes strongly negative correlations large distances, which its labeling rule classifies as perceptually distinct.
“the larger the absolute value of 𝜌, the higher perceptual similarity of the two videos.”
“If the distance is not smaller than 𝑡 𝑟 , the label is assigned as -1, which means the two videos are predicted as perceptual distinct.”
Cites
Rethinking the Inception Architecture for Computer Vision
Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens, J., & Wojna, Z. (2016) · 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) · doi:10.1109/cvpr.2016.308 · Cited by 32,138 · cited 1× · see citing works
“Inspired by the 2D InceptionNet [20], we design an improved Inception block, which can not only extract the spatial information of the frame, but also capture the long-term dependencies of frame sequence, as shown in Fig.”
MethodThe Unreasonable Effectiveness of Deep Features as a Perceptual Metric
Zhang, R., Isola, P., Efros, A. A., Shechtman, E., & Wang, O. (2018) · 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition · doi:10.1109/cvpr.2018.00068 · Cited by 14,623 · cited 1× · see citing works
“…by identifying the object class in the image or video corresponding to the query one, the perceptual robust hashing concerns more about the perceptual features [19] in the image or video to identify the same image/video under the condition of different content-preserving manipulations, see Fig.”
BackgroundA Closer Look at Spatiotemporal Convolutions for Action Recognition
Tran, D., Wang, H., Torresani, L., Ray, J., LeCun, Y., & Paluri, M. (2018) · IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings - IEEE Computer Society Conference on Computer Vision and Pattern Recognition/Proceedings · doi:10.1109/cvpr.2018.00675 · Cited by 3,672 · cited 1× · see citing works
“In order to verify the effectiveness of these two strategies, we replaced the former with R(2+1)D [26], and maintained the other parts.”
A Closer Look at Spatiotemporal Convolutions for Action Recognition
Tran, D., Wang, H., Torresani, L., Ray, J., LeCun, Y., & Paluri, M. (2017) · arXiv · doi:10.48550/arxiv.1711.11248 · Cited by 3,664 · cited 1× · see citing works
“In order to verify the effectiveness of these two strategies, we replaced the former with R(2+1)D [26], and maintained the other parts.”
The Unreasonable Effectiveness of Deep Features as a Perceptual MetricPreprint
Zhang, R., Isola, P., Efros, A. A., et al. (2018) · arXiv (Cornell University) · doi:10.48550/arxiv.1801.03924 · Cited by 810 · cited 1× · see citing works
“…by identifying the object class in the image or video corresponding to the query one, the perceptual robust hashing concerns more about the perceptual features [19] in the image or video to identify the same image/video under the condition of different content-preserving manipulations, see Fig.”
BackgroundVideo Summarization Using Deep Neural Networks: A Survey
Apostolidis, E., Adamantidou, E., Metsai, A. I., Mezaris, V., & Patras, I. (2021) · Proceedings of the IEEE · doi:10.1109/jproc.2021.3117472 · Cited by 366 · see citing works
A Short Note on the Kinetics-700-2020 Human Action Dataset
Smaira, L., Carreira, J., Noland, E., Clancy, E., Wu, A., & Zisserman, A. (2020) · arXiv · doi:10.48550/arxiv.2010.10864 · Cited by 334 · cited 1× · see citing works
“To demonstrate the generalization ability, we randomly selected videos from Kinectics [21] and TRECVID [22] as training data and testing data.”
MethodA Survey on Deep Hashing Methods
Luo, X., Wang, H., Wu, D., Chen, C., Deng, M., Huang, J., & Hua, X.-S. (2023) · ACM Transactions on Knowledge Discovery from Data · doi:10.1145/3532624 · Cited by 241 · cited 1× · see citing works
“Numerous studies [14, 15] have shown that deep learning techniques can effectively extract the features of videos as long as a proper model and loss function are designed.”
BackgroundSelf-citationRecent developments of content-based image retrieval (CBIR)
Li, X., Yang, J., & Ma, J. (2021) · Neurocomputing · doi:10.1016/j.neucom.2020.07.139 · Cited by 208 · cited 1× · see citing works
“Unlike the content-based image/video retrieval [17, 18], which mainly focuses on semantic level retrieval by identifying the object class in the image or video corresponding to the query one, the perceptual robust hashing concerns more about the perceptual features [19] in the image or video to…”
MethodRobust video fingerprinting for content-based video identification
Lee, S., & Yoo, C. D. (2008) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2008.920739 · Cited by 183 · cited 1× · see citing works
“A robust video hashing scheme for content authentication was proposed by Lee and Yoo [8], and this scheme constructed a hash sequence by segmenting each video frame and calculating the centroid of gradient orientations (CGO), which was sensitive to frame rotation and cropping.”
MethodPerceptual hashing for image authentication: A survey
Du, L., Ho, A. T. S., & Cong, R. (2019) · Signal Processing Image Communication · doi:10.1016/j.image.2019.115713 · Cited by 141 · see citing works
Qin, C., Liu, E., Feng, G., & Zhang, X. (2021) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2020.3047142 · Cited by 78 · cited 4× · see citing works
“Note that, the lengths of hash code for [23], [24], [25] and [15] were defaultly set as 96 bits, 100 bits, 256 bits and 75 bits, respectively, and our scheme was 64 bits.”
BackgroundMethodInfluentialSelf-citationYu, M., Tang, Z., Zhang, X., Zhong, B., & Zhang, X. (2022) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2022.3190273 · Cited by 62 · cited 1× · see citing works
“…fingerprint of multimedia data, known as hash code, and can be used in the fields of copy detection [2], content authentication [3], tampering detection [4], etc. Inspired by several excellent image hashing schemes [5–7], some researchers extended the image hashing directly to the video hashing.”
MethodSelf-citationLearning spatial-temporal features for video copy detection by the combination of CNN and RNN
Hu, Y., & Lu, X. (2018) · Journal of Visual Communication and Image Representation · doi:10.1016/j.jvcir.2018.05.013 · Cited by 57 · cited 3× · see citing works
“Note that, the lengths of hash code for [23], [24], [25] and [15] were defaultly set as 96 bits, 100 bits, 256 bits and 75 bits, respectively, and our scheme was 64 bits.”
MethodPerceptual Video Hashing for Content Identification and Authentication
Khelifi, F., & Bouridane, A. (2019) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2017.2776159 · Cited by 56 · cited 1× · see citing works
“The technique of perceptual hashing [1] can map multimedia data into a short sequence, which is equivalent to the fingerprint of multimedia data, known as hash code, and can be used in the fields of copy detection [2], content authentication [3], tampering detection [4], etc.”
MethodRobust and Secure Image Fingerprinting Learned by Neural Network
Li, Y., Wang, D., & Tang, L. (2020) · IEEE Transactions on Circuits and Systems for Video Technology · doi:10.1109/tcsvt.2019.2890966 · Cited by 52 · see citing works
A robust hashing algorithm based on SURF for video copy detection
Yang, G., Chen, N., & Jiang, Q. (2012) · Computers & Security · doi:10.1016/j.cose.2011.11.004 · Cited by 48 · cited 2× · see citing works
“[2] divided the video into frames, calculated the speed up robust feature (SURF) of each frame, and used the Hilbert curve to count the number of feature points to obtain the hash code.”
BackgroundMethodAnuranji, R., & Srimathi, H. (2020) · Digital Signal Processing · doi:10.1016/j.dsp.2020.102729 · Cited by 40 · cited 4× · see citing works
“Note that, the lengths of hash code for [23], [24], [25] and [15] were defaultly set as 96 bits, 100 bits, 256 bits and 75 bits, respectively, and our scheme was 64 bits.”
BackgroundMethodInfluentialUnified Performance Evaluation Method for Perceptual Image Hashing
Li, X., Qin, C., Wang, Z., Qian, Z., & Zhang, X. (2022) · IEEE Transactions on Information Forensics and Security · doi:10.1109/tifs.2022.3161149 · Cited by 32 · cited 1× · see citing works
“…fingerprint of multimedia data, known as hash code, and can be used in the fields of copy detection [2], content authentication [3], tampering detection [4], etc. Inspired by several excellent image hashing schemes [5–7], some researchers extended the image hashing directly to the video hashing.”
MethodSelf-citationUnsupervised Video Representation Learning by Bidirectional Feature Prediction
Behrmann, N., Gall, J., & Noroozi, M. (2021) · 2021 IEEE Winter Conference on Applications of Computer Vision (WACV) · doi:10.1109/wacv48630.2021.00171 · Cited by 28 · cited 1× · see citing works
“But, current perceptual hashing schemes often require manual annotation of video samples and their similarity relationships, and these methods may not have enough generalization capability to deal with massive video and emerging video manipulations [16].”
BackgroundA multi-embedding neural model for incident video retrieval
Chiang, T.-H., Tseng, Y., & Tseng, Y. (2022) · Pattern Recognition · doi:10.1016/j.patcog.2022.108807 · Cited by 24 · see citing works
Visual Attention Based Temporally Weighting Method for Video Hashing
Liu, X., Sun, J., & Liu, J. (2013) · IEEE Signal Processing Letters · doi:10.1109/lsp.2013.2287006 · Cited by 22 · cited 1× · see citing works
“However, unlike image hashing, video hashing encounters great challenges because videos contain spatio-temporal information [9, 10], such as human actions, background transformation and some high-level perceptual features, which make it difficult to capture the key information of videos.”
BackgroundAttention-Based Deep Metric Learning for Near-Duplicate Video Retrieval
Wang, K.-H., Cheng, C.-C., Chen, Y.-L., Song, Y., & Lai, S.-H. (2021) · 2020 25th International Conference on Pattern Recognition (ICPR) · doi:10.1109/icpr48806.2021.9412710 · Cited by 19 · cited 3× · see citing works
“Note that, the lengths of hash code for [23], [24], [25] and [15] were defaultly set as 96 bits, 100 bits, 256 bits and 75 bits, respectively, and our scheme was 64 bits.”
MethodAwad, G., Butt, A. A., Curtis, K., Fiscus, J., Godil, A., Lee, Y., Delgado, A., Zhang, J., Godard, E., Chocot, B., et al. (2021) · arXiv · doi:10.48550/arxiv.2104.13473 · Cited by 18 · cited 1× · see citing works
“To demonstrate the generalization ability, we randomly selected videos from Kinectics [21] and TRECVID [22] as training data and testing data.”
MethodPerceptual video hashing using 3D-radial projection technique
Sandeep, R., Sharma, S., & Bora, P. K. (2017) · 2017 Fourth International Conference on Signal Processing, Communication and Networking (ICSCN) · doi:10.1109/icscn.2017.8085727 · Cited by 12 · cited 1× · see citing works
“Sandeep et al. [13] extended the two-dimensional radial projection technique (RPT) for image hashing to three dimensions by calculating the variance of the luminance values of the projected pixels for each sub-tensor, and then applied DCT to generate a video hash sequence.”
MethodPerceptual hashing in the 3D-DWT domain
Saikia, N. (2015) · doi:10.1109/icgciot.2015.7380552 · Cited by 11 · cited 1× · see citing works
“Saikia [12] performed 3D-DWT on the video and extracted the coefficients from the low-pass sub-bands, and then conducted DCT on these coefficients to obtain the hash code.”
MethodSpatio-temporal digital video hashing using edge orientation histogram and discrete cosine transform
Setyawan, I., & Timotius, I. K. (2014) · 2014 International Conference on Information Technology Systems and Innovation (ICITSI) · doi:10.1109/icitsi.2014.7048247 · Cited by 10 · cited 1× · see citing works
“[11] proposed a spatio-temporal domain hashing scheme based on edge orientation histogram (EOH) and discrete cosine transform (DCT).”
MethodDetection of Malicious Video Modifications using Perceptual Video Hashing
Sandeep, R., & Bora, P. K. (2020) · 2020 5th International Conference on Computing, Communication and Security (ICCCS) · doi:10.1109/icccs49678.2020.9277177 · Cited by 7 · cited 1× · see citing works
“The technique of perceptual hashing [1] can map multimedia data into a short sequence, which is equivalent to the fingerprint of multimedia data, known as hash code, and can be used in the fields of copy detection [2], content authentication [3], tampering detection [4], etc.”
Method
Flagged first, then by citations
3 of the 31 references couldn’t be matched to a known paper.
Cited by
Sorted by citations
A Survey of Perceptual Hashing for Multimedia
Zhou, Y., Li, X., Xiong, C., Yao, H., & Qin, C. (2025) · ACM Transactions on Multimedia Computing Communications and Applications · doi:10.1145/3727880 · Cited by 12 · see citing works
“Zhao 𝑒𝑡 𝑎𝑙 .”
Self-citation
- Authors
- Zhao, G., Qin, C., Luo, X., Zhang, X., & Chang, C.-C.
- Affiliations
- University of Shanghai for Science and Technology (CN)
- State Key Laboratory of Mathematical Engineering and Advanced Computing, Zhengzhou, China
- Shanghai University (CN)
- Feng Chia University (TW)
- Venue
- Proceedings of the 2023 ACM Workshop on Information Hiding and Multimedia Security
- Publisher
- ACM
- Published
- 2023-06-28
- Pages
- 13-18
- OpenAlex
- W4381187965
- Semantic Scholar
- 596821374416ef516e6b7432e889cb700bccdcb8
- Cited by
- 1
- References
- 31
- Citations by year
- 2025: 1
- FWCI
- 0.11
- Type
- conference paper
- Language
- en
- Open access
- bronze
- doi.org
- Locations
- ACM Workshop on Information Hiding and Multimedia Security (IH&MMSec) (publishedVersion)
- Topics
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Digital Media Forensic Detection
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Video Analysis and Summarization
- Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Advanced Image and Video Retrieval Techniques
- Keywords
- video copy detection, unsupervised hashing, multi-scale feature fusion, 3D convolutional neural networks, pseudo-label generation
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-11.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
