Table 6 Performance impact of filters and histogram equalization on the proposed framework across multiple deep learning architectures for validation set (Results in bold highlight the best accuracy value in percentage for each network architecture).

From: A multi-filter deep transfer learning framework for image-based autism spectrum disorder detection

Filters (\(\mathcal {F}\)) / Structure

AffectNet

AlexNet

ResNet-50

ViTSwin

VGG16

VGG19

ViT

ViTFER

None

82.00

82.00

72.00

81.00

70.00

68.00

81.00

80.00

high

83.17

75.00

61.17

81.00

65.00

63.50

80.00

80.17

smooth

83.67

82.00

78.00

86.00

74.00

71.00

81.00

86.00

original

83.83

82.00

79.00

84.33

72.00

70.00

81.00

82.00

histeq_high

81.50

76.00

61.00

81.50

66.00

67.00

79.00

83.00

histeq_smooth

82.00

79.00

78.00

82.50

74.00

74.00

83.00

80.17

histeq_original

83.33

81.00

71.00

81.50

73.00

71.00

83.00

80.83

high+histeq_high

80.50

73.83

60.33

81.17

66.00

66.00

80.00

81.00

original+histeq_high

85.83

81.00

71.00

85.17

73.00

70.00

80.67

82.00

original+smooth+high

84.33

80.00

74.00

86.17

71.00

72.00

81.17

84.00

smooth+histeq_smooth

81.17

78.00

80.00

86.00

77.00

71.67

83.00

84.00

original+histeq_smooth

83.00

79.00

76.00

85.50

76.00

71.50

84.00

84.00

original+histeq_original

82.17

76.17

74.00

84.67

71.00

69.67

84.00

82.00

histeq_original+smooth+high

85.00

81.00

73.00

87.00

72.00

73.00

82.00

83.00

histeq_original+histeq_high

86.00

78.00

70.00

83.83

73.00

71.00

83.00

84.00

histeq_original+histeq_smooth

82.00

76.00

74.00

83.00

75.00

72.00

85.00

80.50

original+histeq_smooth+histeq_high

85.00

80.00

73.00

86.33

71.00

72.00

82.00

81.83

histeq_original+histeq_smooth+histeq_high

86.00

81.00

72.00

84.67

75.00

73.00

82.00

85.00

original+smooth+high+histeq_original+histeq_smooth+histeq_high

87.00

79.00

74.00

86.50

71.00

73.00

81.00

82.33

  1. Significant values are in bold.