Acceso abierto

Research on Multimodal Image Tampering Detection and Counterfeit Image Recognition Techniques under Deep Learning Framework

Meijing Zhang

Meijing Zhang

Key University Laboratory of Fujian Province Digitalized Cyber Crime Supervision Prevention Control, Fujian Police CollegeFuzhou, China
Buscar este autor en
Sciendo|Google Scholar
Zhang, Meijing
, 
Qiang Ding

Qiang Ding

Collaborative Innovation Research Center of Intelligent Policing, Fujian Police CollegeFuzhou, China
Buscar este autor en
Sciendo|Google Scholar
Ding, Qiang
 y 
Xiafei Yan

Xiafei Yan

Department of Computer and Information Security Management, Fujian Police CollegeFuzhou, China
Buscar este autor en
Sciendo|Google Scholar
Yan, Xiafei
  
03 feb 2025
Research on Multimodal Image Tampering Detection and Counterfeit Image Recognition Techniques under Deep Learning Framework's Cover Image
Acerca de este artículo
Artículo anterior
Artículo siguiente
Cite
Descargar portada
Publicado en línea: 03 feb 2025
Recibido: 12 sept 2024
Aceptado: 02 ene 2025
DOI: https://doi.org/10.2478/amns-2025-0018
Palabras clave
© 2025 Meijing Zhang et al., published by Sciendo
This work is licensed under the Creative Commons Attribution 4.0 International License.

Figure 1.

Faking image data processing flow chart
Faking image data processing flow chart

Figure 2.

Structure of the proposed model
Structure of the proposed model

Figure 3.

Workflow of the proposed model
Workflow of the proposed model

Figure 4.

Comparison of model training efficiency
Comparison of model training efficiency

Comparison of the average recognition accuracy

Method Splicing Copy movement Removal Average
MFCN 0.968 0.968
BusterNet 0.941 0.941
RGB-N Net 0.937 0.934 0.913 0.928
YCrCb-N Net 0.951 0.938 0.925 0.938
RGB-ImroveN Net 0.953 0.951 0.932 0.945
Our model 0.964 0.961 0.943 0.956

Robust analysis of common post-processing attacks(NIST16)

Operation Our model MVSSNet IF-OSN
F1 AUC IoU F1 AUC IoU F1 AUC IoU
Control (numerous increases) 0.82 0.93 0.85 0.75 0.95 0.61 0.89 0.98 0.81
Zooming(0.75×) 0.83 0.99 0.86 0.68 0.91 0.62 0.88 0.98 0.84
Zooming(0.35×) 0.90 0.98 0.86 0.57 0.91 0.47 0.81 0.98 0.78
Gaussian blur(4) 0.83 0.99 0.85 0.68 0.92 0.62 0.91 0.95 0.80
Gaussian blur(16) 0.86 0.99 0.77 0.39 0.83 0.30 0.78 0.94 0.80
Gaussian noise(4) 0.80 0.94 0.78 0.59 0.83 0.45 0.82 0.98 0.74
Gaussian noise(16) 0.23 0.85 0.21 0.15 0.70 0.18 0.13 0.65 0.14
JPEG compression(120) 0.89 0.99 0.84 0.73 0.95 0.64 0.94 0.98 0.83
JPEG compression(90) 0.87 0.98 0.79 0.72 0.97 0.62 0.89 0.99 0.86
JPEG compression(80) 0.85 0.99 0.84 0.75 0.92 0.64 0.93 0.96 0.81
JPEG compression(60) 0.92 0.96 0.84 0.70 0.94 0.66 0.74 0.95 0.81

Comparison of F1 scores on two standard data sets

Method CASIA19 NISIT21
MFCN 0.787 0.783
BusterNet 0.781 0.754
RGB-N Net 0.783 0.782
YCrCb-N Net 0.795 0.772
RGB-ImroveN Net 0.813 0.804
Our model 0.835 0.818

Robust analysis of common post-processing attacks(IMD2020)

Operation Our model MVSSNet IF-OSN
F1 AUC IoU F1 AUC IoU F1 AUC IoU
Control (numerous increases) 0.70 0.95 0.60 0.35 0.78 0.30 0.61 0.87 0.52
Zooming(0.75×) 0.59 0.92 0.57 0.41 0.81 0.32 0.57 0.82 0.51
Zooming(0.35×) 0.47 0.91 0.45 0.21 0.80 0.23 0.47 0.78 0.30
Gaussian blur(4) 0.56 0.93 0.60 0.38 0.91 0.29 0.55 0.82 0.50
Gaussian blur(16) 0.54 0.94 0.40 0.19 0.91 0.14 0.40 0.85 0.37
Gaussian noise(4) 0.62 0.91 0.50 0.33 0.83 0.35 0.53 0.89 0.35
Gaussian noise(16) 0.27 0.77 0.18 0 0.55 0 0.11 0.69 0.06
JPEG compression(120) 0.63 0.97 0.59 0.29 0.85 0.30 0.59 0.95 0.53
JPEG compression(90) 0.61 0.98 0.52 0.38 0.86 0.27 0.59 0.95 0.45
JPEG compression(80) 0.51 0.93 0.53 0.42 0.93 0.30 0.48 0.83 0.43
JPEG compression(60) 0.63 0.94 0.46 0.44 0.80 0.21 0.63 0.88 0.44

Experimental results

Model Data volume
NIST12=581 CASIA=974 IMD2016=2022
Accuracy Time(s) Accuracy Time(s) Accuracy Time(s)
MFCN 35.29±2.5% 23.2 43.52±1.93% 99.6 45.31±3.93% 302.3
BusterNet 52.37±2.96% 14.5 60.48±2.61% 49.1 64.08±2.87% 159.1
RGB-N Net 73.20±1.68% 27.8 78.18±1.56% 80.4 82.80±1.03% 717.7
YCrCb-N Net 65.03±2.08% 19.9 75.94±2.02% 124.7 80.12±3.88% 421.1
RGB-ImroveN Net 66.15±1.56% 16.7 77.01±2.38% 43.2 81.07±2.78% 164.3
Our method 74.81±1.77% 26.4 80.07±1.76% 72.8 83.92±1.34% 319.9
Idioma:
Inglés
Calendario de la edición:
1 veces al año
Temas de la revista:
Ciencias de la vida, Ciencias de la vida, otros, Matemáticas, Matemáticas aplicadas, Matemáticas generales, Física, Física, otros
RSS Feed de revista