DeepLumina: A Method Based on Deep Features and Luminance Information for Color Texture Classification

Table 5

Accuracy obtained for DeepLumina Method on benchmark texture dataset FMD.

	RGB	DeepLumina - proposed method
Pretrained models	RGB	RGB + Y	RGB + L	RGB + V	RGB + Y
ColorSpaces	RGB [18]	YCbCr	L ^∗ a ^∗ b ^∗	HSV	YIQ
MobileNet + SVM	74.60	87.1	87.30	85.10	85.80
ResNet50 + SVM	81.60	88.45	89.70	87.85	88.80
ResNet101 + SVM	81.40	89.46	89.65	89.20	90.15
DenseNet201 + SVM	80.75	88.83	87.38	87.46	89.17
AlexNet + SVM	64.30	70.02	70.05	69.50	71.50
VGG19 + SVM	78.10	80.40	81.65	79.35	81.68
Inceptionv3 +  SVM	76.60	88.70	88.15	87.55	89.80
InceptionResNetv2 + SVM	82.20	88.75	90.01	90.03	90.05

The best values obtained for DeepLumina on the FMD dataset are indicated in bold.