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

Table 4

Accuracy obtained for DeepLumina on benchmark texture dataset DTD.

	RGB	Proposed method - DeepLumina
Pretrained models	RGB	RGB + Y	RGB + L	RGB + V	RGB + Y
ColorSpaces	RGB18	YCbCr	L^∗a^∗b^∗	HSV	YIQ
MobileNet + SVM	61.37	66.64	67.31	66.14	68.15
ResNet50 + SVM	67.23	72.47	72.10	70.77	73.01
ResNet101 + SVM	66.92	72.50	72.31	72.43	73.63
DenseNet201 + SVM	65.12	70.66	70.64	68.35	71.37
AlexNet + SVM	46.35	49.75	49.42	48.71	49.84
VGG19 + SVM	55.78	58.60	58.86	58.73	59.53
Inceptionv3 +  SVM	65.20	70.04	71.30	69.81	71.18
InceptionResNetv2 + SVM	65.94	70.78	71.29	70.85	71.39

Best values are shown in bold and they are obtained for the proposed Method DeepLumina for Luminance from the YIQ color model.