Video Fingerprint Algorithm Based on Spatio-Temporal Deep Neural Network

Wang Dongdong; Li Yuenan

doi:10.3788/LOP55.011006

Journals >Laser & Optoelectronics Progress >Volume 55 >Issue 1 >Page 11006 > Article

Laser & Optoelectronics Progress
Vol. 55, Issue 1, 11006 (2018)

Video Fingerprint Algorithm Based on Spatio-Temporal Deep Neural Network

Wang Dongdong^* and Li Yuenan

Author Affiliations

School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China

show less

DOI: 10.3788/LOP55.011006 Cite this Article Set citation alerts

Wang Dongdong, Li Yuenan. Video Fingerprint Algorithm Based on Spatio-Temporal Deep Neural Network[J]. Laser & Optoelectronics Progress, 2018, 55(1): 11006 Copy Citation Text

show less

Fig. 1. Architecture of SFEN

Download full size

Fig. 2. Diagram of DAE

Download full size

Fig. 3. Training of the i+1^th DAE

Download full size

Fig. 4. Diagram of the whole network

Download full size

Fig. 5. Structure of a basic LSTM unit

Download full size

Fig. 6. LSTM network

Download full size

Fig. 7. Fine-tuning of the LSTM network

Download full size

Fig. 8. Comparison of ROC curves

Download full size

Distortion	Description
Compression	Encoder: XVid, frame rate: 25 frame/s, bit rate: 256 kb/s, fixed resolution: 480×320
Median filtering	Filter size ∈[10,20]
Gaussian noise	Zero mean, variance ∈[0.1,0.5,1]
Rotation+croping	Pixel ∈[2,5,10]
Histogram equalization	Number of gray levels ∈[16,32,64]
Frame dropping	Delete 25% frames then linearly interpolate
Frame resizing	Ratio ∈[0.2,4]
Joint spatio-temporal distortion 1	Combine median filtering (filter size is 10) and frame dropping (25%)
Joint spatio-temporal distortion 2	Combine rotation (pixel is 5) and frame dropping (25%)

Table 1. Distortion types and parameters setting

Distortion	Proposedalgorithm	CRBMalgorithm	CGOalgorithm	SGMalgorithm	3D-DCTalgorithm	Rashalgorithm
Compression	0.966	0.907	0.801	0.929	0.889	0.845
Filtering	0.998	0.994	0.887	0.978	0.991	0.966
Noise	0.999	0.999	0.813	0.890	0.896	0.624
Rotation	0.999	0.988	0.534	0.939	0.838	0.950
Equalization	0.996	0.980	0.839	0.836	0.851	0.519
Dropping	0.999	0.999	0.996	0.994	0.999	0.999
Resizing	0.999	0.999	0.903	0.994	0.998	0.948
Joint distortion 1	0.998	0.996	0.941	0.985	0.997	0.981
Joint distortion 2	0.999	0.993	0.516	0.968	0.888	0.965
Overall	0.995	0.982	0.783	0.915	0.894	0.793

Table 2. Comparison of the F₁ scores

Wang Dongdong, Li Yuenan. Video Fingerprint Algorithm Based on Spatio-Temporal Deep Neural Network[J]. Laser & Optoelectronics Progress, 2018, 55(1): 11006

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information