Skip to main content

Table 6 Average classification accuracy for patch size 256 × 256

From: Detection of shifted double JPEG compression by an adaptive DCT coefficient model

QF1

QF2

 

50

60

70

80

90

50

53.24/58.51/59.75

56.39/66.82/64.12

65.63/78.59/71.01

75.82/89.20/86.18

84.39/94.19/93.59

52.84/72.78

57.54/86.17

67.88/95.66

77.05/98.54

86.72/99.81

60

50.91/53.52/52.40

55.02/60.36/57.57

58.23/65.21/63.58

67.78/78.18/71.37

80.62/89.98/88.14

50.08/60.94

54.81/81.80

60.39/84.75

69.72/98.00

81.64/99.78

70

50.81/51.82/49.92

50.68/53.94/52.15

52.14/57.31/57.99

59.04/68.23/69.21

74.18/84.12/82.61

49.97/50.13

50.00/59.53

53.63/75.09

61.30/97.45

75.50/98.82

80

49.62/51.80/51.42

49.39/50.88/49.50

50.04/50.48/50.31

53.55/58.32/59.16

64.82/72.41/71.58

50.53/50.47

49.38/50.17

50.71/60.92

52.89/77.11

64.57/96.87

90

49.99/50.38/50.42

50.72/50.56/49.39

50.02/50.24/49.78

50.03/52.93/51.13

54.39/54.27/53.67

 

50.26/49.34

50.99/50.27

49.67/51.16

49.66/60.97

54.85/75.09

  1. The accuracies a1/a2/a3/a4/a5 in the table are obtained by Luo's (a1), Bianchi I's (a2), Bianchi II's (a3), Chen's (a4), and the proposed method (a5), respectively. The best accuracies among the five methods investigated are boldfaced. Note that an accuracy of 50 is equivalent to random guess.