Study on TCM Tongue Image Segmentation Model Based on Convolutional Neural Network Fused with Superpixel

Table 3

Segmentation results after fusion with superpixel (mean ± standard deviation).

Algorithm	PA	MPA	MIoU	FWIoU
FCN	0.8856 ± 0.0099	0.8959 ± 0.0079	0.7957 ± 0.0153	0.7988 ± 0.0153
FCN + superpixel	0.8954 ± 0.0079	0.9030 ± 0.0070	0.8108 ± 0.0128	0.8143 ± 0.0126
DeepLab	0.8828 ± 0.0070	0.8907 ± 0.0064	0.7910 ± 0.0115	0.7946 ± 0.0110
DeepLab + superpixel	0.8921 ± 0.0049	0.8990 ± 0.0050	0.8056 ± 0.0082	0.8093 ± 0.0078
UrNet	0.9130 ± 0.0039	0.9159 ± 0.0046	0.8393 ± 0.0065	0.8430 ± 0.0065
SpurNet (UrNet + superpixel)	0.9145 ± 0.0043	0.9168 ± 0.0048	0.8417 ± 0.0072	0.8454 ± 0.0072

The best result on each metric is shown in bold face.