Hindawi Publishing Corporation

Evidence-Based Complementary and Alternative Medicine

Volume 2013, Article ID 264742, 11 pages
http://dx.doi.org/10.1155/2013/264742

Research Article
Tongue Color Analysis for Medical Application

Bob Zhang,1,2 Xingzheng Wang,2 Jane You,2 and David Zhang2

1
Department of Electrical & Computer Engineering, Carnegie Mellon University, Pittsburgh, PA 15213, USA
2
Biometrics Research Center, Department of Computing, The Hong Kong Polytechnic University, Hong Kong

Correspondence should be addressed to Bob Zhang; [email protected]

Received 6 December 2012; Accepted 3 March 2013

Academic Editor: Jong Yeol Kim

Copyright © 2013 Bob Zhang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

An in-depth systematic tongue color analysis system for medical applications is proposed. Using the tongue color gamut, tongue
foreground pixels are first extracted and assigned to one of 12 colors representing this gamut. The ratio of each color for the entire
image is calculated and forms a tongue color feature vector. Experimenting on a large dataset consisting of 143 Healthy and 902
Disease (13 groups of more than 10 samples and one miscellaneous group), a given tongue sample can be classified into one of these
two classes with an average accuracy of 91.99%. Further testing showed that Disease samples can be split into three clusters, and
within each cluster most if not all the illnesses are distinguished from one another. In total 11 illnesses have a classification rate
greater than 70%. This demonstrates a relationship between the state of the human body and its tongue color.

1. Introduction using tongue images. 798 tongue samples were captured from
399 patients (two samples from each person), consisting
The human tongue contains numerous features that can be of common illnesses such as diabetes, pancreatitis, and
used to diagnose disease, with color features being the most hypertension, along with 114 images from tongues affected
important [1–4]. Traditionally, medical practitioners would by appendicitis. The samples were captured using a specially
examine these color features based on years of experience [5– designed device consisting of a 3-CCD digital camera, two
11]. However, ambiguity and subjectivity are always accompa- D65 lights and calibrated with a white color plate. Four color
nied with their diagnostic result. To remove these qualitative spaces (RGB, CIExyY, CIELUV, and CIELAB) were evaluated
aspects, tongue color analysis can be objectively analyzed to find the best combination. In their work they reported
through its color features, which offers a new way to diagnose a correct classification of appendicitis to be 66.67%. Pan-
disease, one that minimizes the physical harm inflicted to creatitis (29 samples) and appendicitis (53 samples) tongue
patients (compared with other medical examinations). images again appeared in [14], albeit this time with 56 normal
A literature review on this topic revealed only a few samples. These images were captured with a device designed
papers where color features from the tongue are the main by their research center consisting of a lens, CCD sensor
component used to diagnose disease. Reference [12] used board, two D65 lights, and video frame grabber. Assessing
tongue color along with qualitative and quantitative analysis the same four spaces as [13], the experimental results showed
to examine 207 patients suffering from lung cancer. The that normal and pancreatitis as well as appendicitis and
patients were split into four syndrome groups according pancreatitis can be linearly separated using color features.
to Chinese Medicine, and the CIELAB color model was In [12] the dataset was quite small and only one disease
used for quantitative classification. They reported significant was analyzed. The patients were also diagnosed using Chinese
statistical difference between the four groups when it came to Medicine. As for [13] its dataset was substantially larger
each tongue’s CIELAB value. The correct classification rate of but only appendicitis was classified. The samples in [14]
each group was 69.4%, 54.4%, 72.2%, and 54.4%, respectively. include two illnesses as well as normal, but their sizes are
A color metrics was utilized in [13] to diagnose appendicitis too small to have any statistical significance. Both works in
2 Evidence-Based Complementary and Alternative Medicine

The whole and 98% tongue color gamut in CIExyY color space
0.8

0.7

0.6

0.5

CIEy-axis
0.4
Figure 1: Tongue capture device.
0.3

Table 1: Disease class statistics listing the ID, name, and number of 0.2
samples.
0.1
Disease Disease name No. of
ID samples −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
1 Chronic Kidney Disease 141 CIEx-axis

2 Diabetes 69 Figure 2: A color gamut in the CIExyY color space depicting the
3 Nephritis 10 tongue color gamut inside the red boundary. Furthermore, 98% of
Hypertension the tongue color gamut can be located within the black boundary.
4 66
5 Verrucous Gastritis 25
6 Pneumonia 10
7 Nephritic Syndrome 10 The 100% and 98% tongue color gamut in CIExyY color space
8 Chronic Cerebral Circulation Insufficiency 14
0.8
9 Upper Respiratory Tract Infection 12
10 Erosive Gastritis 10 0.7
11 Coronary Heart Disease 13
0.6
12 Chronic Bronchitis 11
13 Mixed Hemorrhoid 11 0.5
CIEy-axis

14 Miscellaneous 500 0.4

0.3

[12, 14] did not use any image correction to ensure uniform 0.2
feature extraction and analysis under different operating
conditions. Also, [12–14] used some variation of the CIE color 0.1
space to embody the tongue colors, which may not be very
accurate. Therefore, there is a lack of any work on an in- −0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
depth systematic tongue color analysis system for medical CIEx-axis
applications, one that accurately captures the images and
represents its colors using a tongue color gamut [15]. In Figure 3: The tongue color gamut can be represented using several
this paper such an application is described to address these points by drawing lines from the RGB color space.
problems. Tongue images are first captured using a specifi-
cally designed device with image correction. Afterwards, the The rest of this paper is organized as follows. An intro-
images are segmented [16] with the background removed duction to the tongue image acquisition device and dataset
and tongue foreground remaining. Color features from each used is given in Section 2. Section 3 summarizes the tongue
pixel are then extracted and assigned to 1 of 12 colors color gamut and explains how color features are extracted
symbolizing the tongue color gamut [15]. This produces the using it. In Section 4 classification between the two classes
tongue color feature vector. Experimental results were carried of Healthy and Disease is performed. Following this, illnesses
out on a large scale dataset consisting of 143 Healthy and in the Disease class are classified. Finally, concluding remarks
902 Disease samples (diagnosed using Western Medicine) are made in Section 5.
taken from Guangdong Provincial Hospital of Traditional
Chinese Medicine, Guangdong, China. The Disease class was
composed of 13 specific illnesses (with at least 10 samples in 2. Materials
each group) and one sizeable miscellaneous group (made up
of various illnesses). Classification was performed between The tongue database is composed of 1045 images (one image
the illnesses in addition to Healthy versus Disease. per person) split into 143 Healthy and 902 Disease captured at
Evidence-Based Complementary and Alternative Medicine 3

C (Cyan) R (Red) B (Blue) P (Purple) DR (Deep red) LR (Light red)

LP (Light purple) LB (Light blue) BK (Black) GY (Gray) W (White) Y (Yellow)

Figure 4: 12 colors representing the tongue color gamut with its label on top.

v = (0.001 17.067 0 0 6.004 54.722 0.029 0.205 5.551 4.766 3.711 7.899)

C-0.001% R-17.067% B-0% P-0%

DR-6.004% LR-54.722% LP-0.029% LB-0.205%

BK-5.551% GY-4.766% W-3.711% Y-7.899%

Figure 5: Healthy tongue sample, its tongue color feature vector and corresponding 12-color makeup with most of the pixels classified as LR.
4 Evidence-Based Complementary and Alternative Medicine

v = (0.053 0.002 0.085 0 2.092 0 0.001 0.053 35.648 61.108 0.097 0.853)

C-0.053% R-0.002% B-0.085% P-0%

DR-2.092% LR-0% LP-0.001% LB-0.053%

BK-35.648% GY-61.108% W-0.097% Y-0.853%

Figure 6: Disease tongue sample, its tongue color feature vector and corresponding 12-color makeup with most of the pixels classified as GY.

Guangdong Provincial Hospital of Traditional Chinese Med- 3. Methods

icine, Guangdong, China. The patients’ consent was obtained
according to the Declaration of Helsinki and the Ethical The following section describes how color features are
Committee of the Institution in which the work was per- extracted from tongue images. The tongue color gamut is first
formed approved it. The capture device used was a three-chip summarized in Section 3.1. In Section 3.2, every foreground
CCD camera with 8 bit resolution and two D65 fluorescent tongue pixel is compared to 12 colors representing the tongue
tubes placed symmetrically around the camera in order color gamut and assigned its nearest color. This forms the
to produce a uniform illumination. The images captured color features.
were color corrected [17] to eliminate any noise caused by
variations of illumination and device dependency. This allows 3.1. Tongue Color Gamut. The tongue color gamut [15] rep-
consistent feature extraction and classification in the follow- resents all possible colors that appear on the tongue surface
ing steps. Figure 1 shows the capture device. Healthy samples and exists within the red boundary shown in Figure 2 (CIE-
were verified through a blood test and other experiments. xy chromaticity diagram). Further investigation revealed that
If indicators from the tests fall within a certain range they 98% of the points lie inside the black boundary. To represent
were deemed fit. In the Disease class, samples were collected the tongue color gamut using 12 colors, the RGB color space
from inpatients with illnesses determined by their admission is employed and plotted in Figure 3. On the RG line a point Y
note, diagnosed using Western Medical practices. Inpatients (Yellow) is marked. Between RB a point P (Purple) is marked
suffering from the same illness were grouped together into a and C (Cyan) is marked between GB. The center of the RGB
single class. In total there were 13 ailment groups (with at least color space is calculated and designated as W (White), the
10 samples) and one miscellaneous group containing various first of the 12 colors (see Figure 3). Then, for each R (Red), B
illnesses. A summary of the Disease class breakdown is given (Blue), Y, P, and C point, a straight line is drawn to W. Each
in Table 1. Please note any future reference to a specific illness time these lines intersect the tongue color gamut, a new color
in Table 1 will be made using its Disease ID. is added to represent the 12 colors. This accounts for R, Y, C, B,
Evidence-Based Complementary and Alternative Medicine 5

Healthy
(a)

Disease
(b)

Figure 7: Three typical Healthy (a) and Disease (b) samples.

0.04
0.035
30 0.03
25 0.025
C

20 0.02
GY

15 0.015
10 0.01
5 0.005
0 0
80 80 60 70 80
60 70 80
40 20 40 50 60
40 50 60 Y 0 0 10 20 30 R
Y 20 20 30 40
0 0 10 R
Cluster 1
Cluster 1 Cluster 3
Cluster 2
Figure 9: Plot of Cluster 1 versus Cluster 3.
Figure 8: Plot of Cluster 1 versus Cluster 2.

40
30
and P. LR (Light red), LP (Light purple), and LB (Light blue)
GY

20
are midpoints between lines from the black boundary to W, 10
while DR (Deep red) is selected as no previous point occupies 0
that area. More details about the tongue color gamut can be 80
60 80
found in [15]. GY (Gray) and BK (Black) are not shown in 40 50 60 70
20 30 40
Figure 3 since both belong to grayscale. BK 0 0 10 20 R
The 12 colors representing the tongue color gamut are
extracted from Figure 3 and shown in Figure 4 as a color Cluster 3
Cluster 2
square with its label on top. Correspondingly, its RGB and
CIELAB values are given in Table 2. Figure 10: Plot of Cluster 3 versus Cluster 2.
6 Evidence-Based Complementary and Alternative Medicine

Cluster 1
(a)

Cluster 2
(b)

Cluster 3
(c)

Figure 11: Three typical samples from each cluster.

Figure 12: Three typical samples from Disease 1.

Evidence-Based Complementary and Alternative Medicine 7

Figure 13: Three typical samples from Disease 3.

Figure 14: Three typical samples from Disease 5.

Figure 15: Three typical samples from Disease 6.

Figure 16: Three typical samples from Disease 7.

8 Evidence-Based Complementary and Alternative Medicine

Figure 17: Three typical samples from Disease 8.

Figure 18: Three typical samples from Disease 9.

Figure 19: Three typical samples from Disease 10.

Figure 20: Three typical samples from Disease 11.

Evidence-Based Complementary and Alternative Medicine 9

Figure 21: Three typical samples from Disease 12.

Figure 22: Three typical samples from Disease 13.

3.2. Tongue Color Features. Given a tongue image, segmen- Table 2: RGB and CIELAB values of the 12 colors.
tation is first applied to locate all foreground tongue pixels Color [R G B] [L A B]
[16]. Having located each pixel its corresponding RGB value
C (Cyan) [188 188 185] [76.0693 −0.5580 1.3615]
is extracted and converted to CIELAB [18] by first converting
RBG to CIE𝑋𝑌𝑍 using R (Red) [189 99 91] [52.2540 34.8412 21.3002]
B (Blue) [183 165 180] [69.4695 9.5423 −5.4951]
𝑋 0.4124 0.3576 0.1805 R P (Purple) [226 142 214] [69.4695 42.4732 −23.8880]
[ 𝑌 ] = [0.2126 0.7152 0.0722] [G] (1) DR (Deep red) [136 72 49] [37.8424 24.5503 25.9396]
[ 𝑍 ] [0.0193 0.1192 0.9505] [ B ] LR (Light red) [227 150 147] [69.4695 28.4947 13.3940]
LP (Light purple) [225 173 207] [76.0693 24.3246 −9.7749]
followed by CIE𝑋𝑌𝑍 to CIELAB via
LB (Light blue) [204 183 186] [76.0693 7.8917 0.9885]
𝑌 BK (Black) [107 86 56] [37.8424 3.9632 20.5874]
𝐿∗ = 116𝑓 ( ) − 16, GY (Gray) [163 146 143] [61.6542 5.7160 3.7317]
𝑌0
W (White) [200 167 160] [70.9763 10.9843 8.2952]
𝑋 𝑌 [166 129 93] [56.3164 9.5539 24.4546]
𝑎∗ = 500 [𝑓 ( ) − 𝑓 ( )] , Y (Yellow)
𝑋0 𝑌0
𝑌 𝑍 (2)
𝑏∗ = 200 [𝑓 ( ) − 𝑓 ( )] , V = [𝑐1 , 𝑐2 , 𝑐3 , 𝑐4 , 𝑐5 , 𝑐6 , 𝑐7 , 𝑐8 , 𝑐9 , 𝑐10 , 𝑐11 , 𝑐12 ] and 𝑐𝑖 represents
𝑌0 𝑍0
the sequence of colors in Table 2. As an example, the color
1/3 features of two tongues are shown in visual form (refer to
{𝑥 (𝑥 > 0.008856) ,
where 𝑓 (𝑥) = { 16 Figures 5 and 6) along with its extracted tongue color feature
7.787𝑥 + (𝑥 ≤ 0.008856) . vectors, where the original image is decomposed into one of
{ 116
the 12 colors. Figure 5 is from a Healthy sample and Figure 6
In (2), 𝑋0 , 𝑌0 , and 𝑍0 are the CIE𝑋𝑌𝑍 tristimulus is from a Disease sample. In the Healthy sample the majority
values of the reference white point. The LAB values are of pixels are LR and for Disease it is GY.
then compared to 12 colors from the tongue color gamut The mean colors of Healthy and Disease are displayed
(see Table 2) and assigned the color which is closest to it in Table 3 along with three typical samples from each class
(measured using Euclidean distance). After calculating all shown in Figure 7. Disease tongues have a higher ratio in R,
tongue foreground pixels, the total of each color is summed DR, BK, GY, and Y according to Table 3. On the other hand,
and divided by the number of pixels. This ratio of the LR and W are greater in Healthy. Only 7 colors are listed out
12 colors forms the tongue color feature vector V, where of the 12 as the remaining 5 have ratios less than 1%.
10 Evidence-Based Complementary and Alternative Medicine

Table 3: Mean of the color features for Healthy and Disease.

R DR LR BK GY W Y
Healthy 20.9284 5.6679 33.8483 8.2356 14.5583 7.9166 8.0485
Disease 28.2901 15.5951 11.0277 15.4325 16.2247 2.4990 10.6382

4. Results and Discussion Table 4: Classification result between Healthy versus Disease using
𝑘-NN and SVM.
In this section classification using color features is described.
Classification method Average accuracy
Classification between Healthy versus Disease is first given
in Section 4.1, while illnesses in Disease are classified in k-NN 91.99%
Section 4.2. SVM 91.99%

4.1. Healthy versus Disease Classification. Table 4 shows the Table 5: Distribution of the illnesses within the clusters.
classification rate between Healthy versus Disease on the test
Cluster number Disease group
data. Half the images were randomly selected from either
1 1 13
class to represent the training set and the remaining samples
assigned to the test set. The training data in each class are 2 5 7 8 9 10 12
the mean tongue color features of Healthy and Disease. To 3 2 3 4 6 11
reduce the number of tongue color features, feature selection
with sequential forward search was implemented. Both 𝑘-
NN [19] and SVM [19] using a quadratic kernel were tested Diseases 1 and 13 in Cluster 1 are separable with an
producing the same result as can be seen in Table 4. This average accuracy of 76.08%. In Cluster 2, Disease 7 can
means for 𝑘-NN and SVM the tongue color feature vector be first removed as its classification rate of 93.06% is the
of the training set consisting of Healthy and Disease was highest amongst the six illnesses. Diseases 10, 8, and 9 are
placed in an 𝑛-dimensional space. Each tongue color feature subsequently taken out in that order which leaves illnesses
vector representing the test set was mapped to this space and 5 and 12 (classification rate of 81.45%). Looking at Cluster
classified depending on its classification rule (𝑘-NN or SVM). 3, Disease 6 with a classification rate of 74.05% is initially
removed from the pack. This is followed by Diseases 3 and 11
4.2. Typical Disease Analysis. With Healthy versus Disease leaving 2 and 4 which produced the lowest classification result
separated the next step is to examine whether certain illnesses of 54.41%. Table 8 summarizes this result. Diseases 1, 3, 5, 6, 7,
within the Disease class can be distinguished from one 8, 9, 10, 11, 12, and 13 achieved an average accuracy greater than
another. All 13 illnesses were grouped into three clusters by 70% and therefore deemed successfully classified. Typical
FCM [19], with Table 5 illustrating which cluster each illness samples of the successfully classified illnesses are shown in
belongs to. The mean tongue color features of each cluster Figures 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, and 22.
are shown in Table 6. R, DR, and LR are greater in Cluster 3. As part of the future work we plan on returning to Guang-
Cluster 2 has higher concentrations of GY, BK, and W, while dong Provincial Hospital of Traditional Chinese Medicine
Y is more significant in Cluster 1. and collect more diseased tongue images. Color features
Table 7 shows the classification rate of the three clusters (discussed in Section 3.2) will be extracted from these new
calculated in groups of two. In each case the two clusters images before combining it with the previous batch. The
in question are clearly separable as seen in this table and experimental results in the form of Healthy versus Disease
Figures 8, 9, and 10. Three typical samples from each cluster classification and typical disease analysis will be recalculated
are depicted in Figure 11. From a visual perspective the tongue in order to further validate its statistical accuracy.
color features in each cluster are quite different compared to
the rest. 5. Conclusion
Next, each cluster is examined one by one to deter-
mine whether illnesses within it can be classified. This is Given a tongue image the tongue color analysis system is able
accomplished by comparing illnesses inside the cluster and to first distinguish Healthy versus Disease with an average
removing the illness with the highest classification. The accuracy of 91.99%. If the image is from Disease it is further
process is repeated until all illnesses have been classified. assigned to one of three clusters. From these clusters 11
The same experimental setup described in Section 4.1 was illnesses can be successfully classified given a classification
applied, where half the images are randomly selected for rate of at least 70%. The proposed method uses a special
training and test sets. Both 𝑘-NN and SVM were used as the capture device with image correction and extracts a tongue
classifiers along with sequential forward search for feature color feature vector from each image. This vector consists
selection. An illness is considered successfully classified if its of 12 color ratios calculated with the tongue color gamut to
average accuracy is greater than or equal to 70%. The average better characterize each foreground tongue pixel. Testing was
accuracies stated in the following paragraph represents only carried out on a large dataset collected from Guangdong,
SVM. For a complete list of the results please refer to Table 8. China, consisting of 143 Healthy and 902 Disease images
Evidence-Based Complementary and Alternative Medicine 11

Table 6: Mean tongue color features of the three clusters.

Cluster number R DR LR BK GY W Y
1 21.561 13.972 12.265 12.535 9.524 3.703 26.191
2 17.116 11.980 9.437 15.574 34.733 4.111 6.539
3 40.736 15.396 15.872 8.232 10.770 1.668 7.015

Table 7: Classification result between the three clusters compared [4] B. Kirschbaum, Atlas of Chinese Tongue Diagnosis, Eastland
in groups of two. Press, Seattle, Wash, USA, 2000.
[5] B. Li, Q. Huang, Y. Lu, S. Chen, R. Liang, and Z. Wang, “A
Cluster comparison Average accuracy
method of classifying tongue colors for traditional chinese
Cluster 1 versus Cluster 2 100% medicine diagnosis based on the CIELAB color space,” in Pro-
Cluster 1 versus Cluster 3 97.75% ceedings of the International Conference on Medical Biometrics,
Cluster 2 versus Cluster 3 99.63% pp. 153–159, 2007.
[6] C. H. Li and P. C. Yuen, “Tongue image matching using color
content,” Pattern Recognition, vol. 35, no. 2, pp. 407–419, 2002.
Table 8: Classification result of the illnesses.
[7] N. M. Li, “The contemporary investigations of computerized
Disease ID Cluster number k-NN SVM tongue diagnosis,” in The Handbook of Chinese Tongue Diagno-
1 1 72.53% 76.08% sis, Shed-Yuan Publishing, 1994.
2 3 54.87% 54.41% [8] G. Maciocia, Tongue Diagnosis in Chinese Medicine, Eastland
3 3 72.97% 75.68% Press, Seattle, Wash, USA, 1995.
4 3 54.87% 54.41% [9] N. Ohta and A. Robertson, Colorimetry: Fundamentals and
Applications, John Wiley & Sons, 2006.
5 2 81.45% 81.45%
[10] B. Pang, D. Zhang, and K. Wang, “Tongue image analysis for
6 3 78.61% 74.05%
appendicitis diagnosis,” Information Sciences, vol. 175, no. 3, pp.
7 2 83.33% 93.06% 160–176, 2005.
8 2 72.02% 83.33% [11] B. Pang, D. Zhang, and K. Wang, “The bi-elliptical deformable
9 2 77.78% 83.33% contour and its application to automated tongue segmentation
10 2 78.71% 87.10% in chinese medicine,” IEEE Transactions on Medical Imaging,
11 3 72.82% 73.56% vol. 24, no. 8, pp. 946–956, 2005.
12 2 81.45% 81.45% [12] W. Su, Z. Y. Xu, Z. Q. Wang, and J. T. Xu, “Objectified study
13 1 72.53% 76.08% on tongue images of patients with lung cancer of different
syndromes,” Chinese Journal of Integrative Medicine, vol. 17, no.
4, pp. 272–276, 2011.
[13] K. Q. Wang, D. Zhang, N. M. Li, and B. Pang, “Tongue diagnosis
(13 specific illnesses with at least 10 samples and a miscella-
based on biometric pattern recognition technology,” in Pattern
neous folder). The experimental results showed that there is a Recognition from Classical to Modern Approaches, S. K. Pal and
relationship between tongue color and the state of the human A. Pal, Eds., pp. 575–598, World Scientific, 1st edition, 2001.
body, which can be used in medical applications to detect [14] X. Wang and D. Zhang, “An optimized tongue image color cor-
various illnesses. This can potentially lead to a new painless rection scheme,” IEEE Transactions on Information Technology
and efficient way to examine patients. in Biomedicine, vol. 14, no. 6, pp. 1355–1364, 2010.
[15] X. Wang and D. Zhang, “Statistical tongue color distribution
Acknowledgments and its application,” in Prococeedings of the International Con-
ference on Computer and Computational Intelligence, 2011.
The work is partially supported by the GRF fund from the [16] D. Zhang, Automated Biometrics: Technologies and Systems,
HKSAR Government, the central fund from Hong Kong Kluwer Academic Publisher, Boston, Mass, USA, 2000.
Polytechnic University, and the NSFC Oversea fund [17] D. Zhang, B. Pang, N. Li, K. Wang, and H. Zhang, “Computer-
(61020106004), China. ized diagnosis from tongue appearance using quantitative fea-
ture classification,” The American Journal of Chinese Medicine,
vol. 33, no. 6, pp. 859–866, 2005.
References [18] H. Z. Zhang, K. Q. Wang, X. S. Jin, and D. Zhang, “SVR
[1] C. C. Chiu, “Development of a computerized tongue diagnosis based color calibration for tongue image,” in Proceedings of the
system,” Biomedical Engineering, vol. 8, no. 4, pp. 342–350, 1996. International Conference on Machine Learning and Cybernetics
(ICMLC ’05), pp. 5065–5070, August 2005.
[2] C. C. Chiu, “A novel approach based on computerized image
analysis for traditional Chinese medical diagnosis of the [19] Y. Zhang, R. Liang, Z. Wang, Y. Fan, and F. Li, “Analysis of the
tongue,” Computer Methods and Programs in Biomedicine, vol. color characteristics of tongue digital images from 884 physical
61, no. 2, pp. 77–89, 2000. examination cases,” Journal of Beijing University of Traditional
Chinese Medicine, vol. 28, pp. 73–75, 2005.
[3] R. O. Duda, P. E. Hart, and D. G. Stork, Pattern Classification,
Wiley & Sons, 2nd edition, 2000.