Table 3 Performance comparison of the proposed passive KB-CMFD method with state-of-the-art CMFD methods on publicly available datasets (best performances are in bold).

From: Image tampering detection using dynamic histogram equalization based LIOP features and novel scaled K-means +  + clustering

Dataset

CMFD methods

P

R

F-measure

CoMoFoD

Clustered-based18

0.7980

0.7490

0.7727

SIFT41

0.7000

0.8750

0.7778

SIFT-IMC-RG43

0.7019

0.8461

0.7673

Passive framework44

0.9660

0.9800

0.9729

FMT45

0.8290

0.7851

0.8064

SURF46

0.6160

0.7098

0.6596

ORB47

0.7692

0.8000

0.7843

KAZE47

0.8800

0.8461

0.8627

AKAZE47

0.8889

0.9600

0.9231

DL-CNN48

0.9794

0.9520

0.9382

Proposed passive KB-CMFD method

0.9962

0.9597

0.9776

MICC-F220

Clustered-based18

0.9050

0.9550

0.9293

FMT45

1.0000

0.5940

0.7453

SURF46

0.8160

0.9273

0.8681

Passive framework44

0.9835

0.9682

0.9757

Level-2 clustering49

0.9576

0.9431

0.9503

BB-CMFD50

0.9251

0.9695

0.9468

ResNet5051

0.8800

0.8400

0.08400

Proposed passive KB-CMFD method

0.9891

0.9789

0.9839

Defecto MSCOCO (synthetic)

D2PRL52

0.9470

0.7670

0.8320

Serial network retrain53

0.8560

0.4980

0.5830

DOA-GAN54

0.6560

0.2690

0.3290

BusterNet retrain55

0.6710

0.5040

0.5440

Proposed passive KB-CMFD method

0.9560

0.9147

0.9349

Back to article page