Content uploaded by Suresh G R
Author content
All content in this area was uploaded by Suresh G R on Oct 22, 2019
Content may be subject to copyright.
I J C T A,
© International Science Press
Texture Analysis of Diabetics Tongue using
Binary Pattern
A.Selvarani* and G.R.Suresh**
Abstract: One of the most important diagnosis methods is tongue analysis. Usually, tongue analysis technique
involves in image processing. But it’s hard to process a tongue image due to its dissimilarities and irregularities.
In this paper, we propose a method to overcome these by using texture analysis technique. This process
comprises of four steps: Image Acquisition, Image segmentation, texture analysis and color analysis. The image
acquisition provides the processing basis. Image segmentation is performed using texture analysis. The
experiment results showed that the tongue is segmented and analyzed rightly.
Keywords: Tongue analysis, image processing, texture analysis, level set algorithm, texture classification
1. INTRODUCTION
Tongue analysis is one of the diagnosis processes in the patient administration. The tongue has an
association with the meridian and interior organs. The tongue has numerous connections and
relationships in the body. The tongue has a special relationship with the Heart. It is therefore extremely
helpful and vital amid an investigation for confirming TCM diagnosis. Any neurotic change in the heart
and lungs corresponds to the tongue tip while that of liver and gall bladder corresponds to the respective
sides of the tongue. The doctor can understand the physical and mental condition of the patient by
examining the features of the tongue. For the analysis of tongue image, we have to extract the shape,
colour and texture feature from the central rectangle area. The framework consists of obtaining the
correct colour and finding coating by filtering and enhancement and detection of pimples and cracks of
the tongue.Our proposed method is a sequential process; results rely on input and output, so the expected
outcome of diagnosis is tongue.
In traditional Chinese medicine, diagnosis using tongue is an important diagnostic method. But it has
a very constrained application because of its subjective and test based nature. In Tongue image
processing, the pictures of the tongue are the elementary components for the diagnosis of different
sickness. For the simplicity of the analysis, the tongue pictures ought to be handled openly and
appropriately.[1] Due to previously mention issues like irregular shape of the tongue, lips, etc., it’s hard
to diagnose the disease with the tongue image processing. Tongue diagnosis is one of the necessary
fields in diagnosing most of the illnesses. Tongue diagnosing has received more consequence among the
experts. It is usually carried out by processing the tongue images, which processing of tongue image is
not an easy task to do. The difficulty strikes because of interference of lip with the tongue, the different
shape of the tongue, the irregular shape of the tongue, etc. [2]
* Research Scholar, Dept. of Electronics and Communication Engineering, Sathyabama University, Jeppiaar Nagar,Rajiv Gandhi Road,
Chennai, India, E-mail: selvaranime2001@gmail.com
** Research Supervisor, Dept. of Electronics & Communication Engineering, Sathyabama University, Jeppiaar Nagar,Rajiv Gandhi Road,
Chennai, India, E-mail:sureshgr@rediffmail.com
2 A.Selvarani and G.R.Suresh
2. Related Works
To find most of the diseases tongue diagnosis is widely used by the experts tongue images are collected
to perform tongue diagnosis, but tongue image processing is not an easy way due to the irregular shape,
disturbance of lips, and different shape for each person. The sequential method was used to process the
tongue imaging in this work. Three processes are carried out first find the form of the tongue, second
colour extraction, crack detection and pimple detection is found then finally LGXP method is
implemented to determine the texture of the tongue image. This process reduces the complexity of
analyzing the tongue image. [2] In traditional Chinese medicine tongue diagnostics is an essential one.
So tongue area from the digital image is extracted to perform automatic tongue diagnostic system. Due
to the weaker edges of the tongue, simple segmentation methods will fail to obtain details efficiently
over the tongue surface. Here we combine two methods namely watershed transform and active contour
model are proposed to perform unique segmentation. The watershed transform achieves initial shape,
and the precise edge is converged using active contour model. These techniques provide us more
efficient results on tongue diagnosis [3]. Colour image matching problem in medical diagnosis is the
theme in this work. . In this paper, a colour mapping technique is used on different metrics in different
colour spaces on tongue image. We consider distance measurement in coordinate space, but not solved
the reflexivity axiom problem. To overcome reflexivity axiom problem, we proposed probabilistically
combined metric and sorted metric methods. Sorted metric will over the reflexivity axiom in coordinate
space, and probabilistically combined metric is used to improve the matching performance in colour
matching technique. The analysis is done on tongue image using the proposed methods for matching and
analyzing the results generated [4]. Tongue diagnosis is a preliminary test in medical diagnosis. By
characterization, the accuracy of tongue diagnosis can be determined. Here, image tongue
characterization is used; digital colour tongue images are acquired using TIAI. The overall accuracy
analysis is increased using colour calibration, quantitative analysis and tongue area segmentation. The
colour tongue and its coating, cracks are characterized to improve the overall accuracy of the analysis.
The accuracy of the as well over 85% using the proposed characterization method [5]. The
characteristics of 12 tongue characteristics are analyzed elaborately. Based on a database of 9000 tongue
images, colorimetric imaging system the tongue is characterized by CIE chromaticity diagram. SVM
algorithm and colour gamut boundary descriptor are used to derive the chromaticity diagram. By this
method, 12 tongue characteristics are obtained. To prove the effectiveness, a new colour feature
extraction is proposed for diagnosis [6].
Texture Analysis of Diabetics Tongue using Binary Pattern
3
3. EXISTING SYSTEM
Image segmentation partitions a digital image into multiple segments. The aim of segmentation is to
simplify and change the presentation of an image into meaningful and easier to analyze. Image
segmentation is used to locate boundaries and objects in images. Image segmentation is the process of
allocating a label to every pixel in an image. Image segmentation is a set of segments that cover the
entire image. Pixels in a region are similar to some characteristic or computed property, such a colour,
intensity, or texture. Adjacent areas are different in the same characteristic(s). After image segmentation
can be used to create 3D constructions with the help of interpolation algorithms. The segmentation of
the image is useful in the medical field, computer vision and satellite imaging. The criteria for image
segmenting are very hard. There is a vast amount of literature available to analyze and understand the
segmentation techniques. The technique in medical image segmentation mainly works on fuzzy-c means
and Otsu’s method after applying a vector median filter, for segmentation and has tried to prove the
hard-wearing of their method. Noise is added to sample image to obtain better results.
4. ALGORITHM
Image segmentation is change into a set of visually distinct and same regions on individual properties.
The aim of segmentation is clearly differentiating background object in an image. Tongue area
segmentation is important because it is used for analysis of tongue images to extract tongue area out. In
tongue segmentation, they are many methods proposed; they are threshold segmentation, Region
growing, watershed, snake and so on. And in that tongue segment is best. The Level-set algorithm is
applied in image segmentation.
4.1. COMPRESSION BASED METHOD
The optimal segmentation is one which maintains the overall segmentation and coding length of the
data. Segmentation is used to compress the image by finding any particular pattern or if there is any
connectivity in image characteristics. Each part of the image is modeled using probability distribution
function that can be achieved by representing each segment by its texture and boundary.
The regions in raw images have a smooth contour that is feasible for boundary encoding. The
continuous coding can be attained in short code by using Huffman code to encode the difference chain
system. Lossy compression is used for texture encoding. The texture modeling is done by using normal
distribution. The aim is to find the segmentation method that will produce shortest code length that can
be achieved by agglomerative clustering method. The number of bits required to encode the image is
acquired using the prior scheme.
4 A.Selvarani and G.R.Suresh
4.2. HISTOGRAM BASED METHOD
Histogram-based methods are very efficient than other image segmentation methods. The histogram is
enumerated from pixels in the image, and the peaks and valleys in the histogram are used to find the
clusters in the image. Grouped images are divided into smaller groups. This operation continues with
smaller clusters up to terminal groups are formed.
4.3. COLOUR HISTOGRAM
An image can be encoded using global colour histogram that is the widely used feature because it is easy
to extract. The distance in the colour histogram is used to determine the distance between two images.
LCH (local colour histogram) can provide spatial information and information depending colour
distribution region. Initially, colour histogram for each block is extracted by segmenting the image into
blocks. The distance is then calculated by comparing the newly formed image histograms and the
original image. The sum of all the distances is used to determine the distance between the two images.
4.3.1. DISADVANTAGES
The histogram-seeking method is not suitable for extracting significant peaks and valleys from an
image. Histogram analysis is adequately proper in the areas where multiple frames are engaged.
Histogram analysis can be applied to single frame images but should use multiple times on the same
image. The results are collected and merged into a single image will produce the current peaks and
valleys of the image. This technique can also use as per pixel basis that will help us to identify the
various colours, in particular, pixel location. This technique will be most suitable for active objects with
static environment. This method is used to generate segmentation in the video tracking process.
4.4. REGION GROWING METHOD
Region growing method is mostly depends on assumption of values in neighbouring pixels. This method
assumes whether the neighbouring pixels have same values if the neighbouring node as the same values
then they are grouped into same cluster in the same way many clusters are created to form a perfect
pixels arrangement. The result of this method is very important in image analysis but influence of the
noise will be more in this method.4connecteness of edges was compared with intensity difference was
performed to generate graph of pixels using Region Merging (SRM) method. Each pixel value generates
a Pixel region. The pixel regions ere compared and grouped into priority queue to find whether the
regions are going to merge are not using statistical analysis.
Seeded region growing method is effective in one region growing method. This method collects a group
of seeds as input along with the image. This seeds are used to perform segmentation of objects in the
input image. Pixel intensity value and mean are taken as the output. The values are compared to assign
to a particular region. This process was continued until all pixels are compared in whole input image.
The segmented image depends on the input seeds, if any noise present in the image will result in poor
Texture Analysis of Diabetics Tongue using Binary Pattern
5
alignment of seeds in the output.
The alternate region growing method is non-seeding growing method. This method does not use seeds to
segment the objects from the image. It uses Image iterations are created to analyze the neighbouring
nodes the process is same as seeding method.
4.5. SEGMENTATION OF SUBLINGUAL VEINS
Sublingual vein inspection will provide exceptional data necessary to monitor health condition of
humans. Diagnosis of portal hypertension and blood stasis can be validated using shape and breadth of
sublingual vein. An image acquired by camera under normal light was used in experimental research
papers for sublingual vein analysis.
5. METHODOLOGY
In Tongue image processing, the pictures of tongue are the elementary components for the diagnosis of
different sickness. For the simplicity of the analysis, the tongue pictures ought to be handled clearly and
appropriately. Due to previously mention issues like irregular shape of the tongue, lips etc., it’s hard to
diagnose disease with the tongue image processing. The principal characteristics of tongue are
considered for image processing. The changes in these characteristics reflect the abnormalities of the
body. With the help this analysis, it is easy to diagnose most of the diseases. For the comprehensive
analysis, detailed images of tongue are used. Now, let us consider some tongue images and the disease
analysis.
5.1. TEXTURE ANALYSIS
The foremost step in image processing is image acquisition as it captures the necessary image and
provides a basis for processing. Then the second step, Image segmentation, is used to partition the image
into a set of visually distinct and similar regions with respect to certain attributes. The image
segmentation is implemented mainly to recognize the object and the background in the image. With this,
the tongue area is extracted out completely. For image segmentation, the Level-set algorithm is used.
The initial position of the tongue is decided by the Hue (H) and value (V) of HSV space. The middle
region of tongue that corresponds to pancreas region is segmented separately. In the next step, Texture
analysis, three principles are used to analyze texture such as statistical, structural and spectral. The
statistical approach yields smooth and coarse texture. Then the structural technique deals with the
grouping of image basics. The image basics include texture description based on parallel lines. The final
principle spectral technique is based on Fourier spectrum. It identifies high energy narrow peaks in the
spectrum. In the fourth step, color analysis, color is extracted from different color spaces. It explains the
position of the color in the color space like RGB, CIELAB, CIEYXY, HSV and CIELUV. RGB is an
6 A.Selvarani and G.R.Suresh
additive color system which is applied to use CRT to display images. It is used in most of the application
and also it is very easy to implement. CIE has three primaries x, y and z. In this y is proportional to the
luminance, x and z as an additional component. CIE is used to equalize amounts of the three primaries
which are required to match the light.
Tongue Image Acquisition
Image Segmentation
Texture Analysis
Colour Analysis
Diagnosis Results
Fig Block diagram of texture analysis
HSV (Hue-saturation-value) is the most cylindrical coordinate representation of points in an RGB.HSV
has discontinuities, which make the system noise sensitive. Hue is the color type and saturation is refer
to the intensity of specific hue. Value refers to the brightness of the color. Finally the analysis between
normal and diseased tongue is based on the color and texture features extracted from a set of tongue
images.
5.1.1. APPLICATIONS
The texture analysis has four major applications domains such as texture classification, texture
segmentation, shape from texture, and texture synthesis.
5.2. TEXTURE CLASSIFICATION
Texture classification assigns texture to some texture classes. The two classifications are supervised and
unsupervised classification. A supervised classifier learns a definition for each texture class.
Unsupervised classification does not require knowledge to distinguish different classes from the textures
given. The last class is semi-supervised. It has partial prior knowledge.
Most of the classification methods involve a two-stage process. In the first stage, feature extraction, each
texture class is classified with respect to feature measures. It identifies and selects that are relevant
features such as rotation, scaling and translation. It finds hard time in designing a universal feature
extractor since it depends on problems. Then classification is the second stage. Here, the classifiers
Texture Analysis of Diabetics Tongue using Binary Pattern
7
determine the input texture classification based on measures of preselected features obtained. It
identifies the selected features and yields texture classes.
The textons create a global texton dictionary, in which each material is denoted by a probability density
function. It is represented in terms of spectrum of texton frequency. This sorts image data into clear
information.
5.3. TEXTURE SEGMENTATION
Texture segmentation distributes an image into an array of unconnected regions based on texture
properties. These results are used in further analysis like object recognition. Segmentation involves in
feature extraction and derives a system to differentiate textures. The supervised texture segmentation
deals with boundary separation by detecting different texture regions.
While the unsupervised segmentation recovers various texture classes before segmenting the image into
regions. It is more expensive than supervised segmentation but it is more flexible.
The segmentation is very crucial in TDs. Region growing, edge detection and other low level techniques
do not segment tongue while the high level image processing technique is proposed.
5.3.1. SHAPE FROM TEXTURE
Analyzing the texture properties of a 2 dimensional tongue image makes it difficult to estimate a 3
dimensional surface shape. Weak isotropy and homogeneity of a texture provides a suggestion of shape.
When the surface is seen from a slant, perspective projection results in texture gradient. So the
perspective transformation gives the surface shape parameters. Hence, through a suitable texture
gradient measure, object shape and depth map can be recovered.
Shape from the texture recovers true surface orientation, reconstructs surface shape. It is also used for
inferring the 3D layout of objects. A plane vanish line is extracted from texture deformation.
5.4. TEXTURE SYNTHESIS
Texture synthesis creates a wide texture from tiny texture samples. The created texture differs from
sample but has identical features of those samples. It can handle boundary condition and neglects
repeated verbatim. It gives an empirical solution to check the texture analysis. The texture synthesis
needs a clear texture description. The reproduction of textures is more difficult.
Figure 4: Synthetic Texture
8 A.Selvarani and G.R.Suresh
6. RESULT ANALYSIS
In this chapter we analyzed the different postures of tongues and compared the difference in the
visualization in the tongue images and also provide the results for color analysis of tongue image,
segmentation image, gray scale image and filter outputs are provided below. This output provides as
brief knowledge about the tongue diagnosis in modern medicine.
COMPARISON OF TONGUE
Normal Tongue
After food Before Food
The above image shows the comparison of the tongues in three categories: Normal tongue, before food
and after food. An image shows the difference in presence of pores and white sediments in these images.
The above image is the output of color analysis. The whiter segments are more visible in the result
image after applying color analysis technique in normal image of tongue. This image shows a normal
tongue and the output result after performing colour analysis.
Segmented image Gray scale image
The above image shows the segmented result of the normal tongue and gray scale output of the same
images.
Texture Analysis of Diabetics Tongue using Binary Pattern
9
The above images show the filter output of normal images.
7. CONCLUSION
The tongue diagnosis of diabetes mellitus based on pathological changes on the surface of the tongue
establishes a relationship between tongue appearance and diseases. The analysis between normal and
diseased tongue is based on the color and texture features extracted from a set of tongue images.
Experiments are implemented and results are promising. The main contribution is to bridge the gap
between tongue signs to western medicine defined diseases. From the result it is showed that the method
we give the appropriate results and is well suited for tongue analysis and clinical application.
REFERENCES
[1].Bo Pang; Zhang, D.; Naimin Li; Kuanquan Wang, "Computerized tongue diagnosis based on
Bayesian networks," in Biomedical Engineering, IEEE Transactions on , vol.51, no.10, pp.1803-1810,
Oct. 2004
[2].M. Dhanalakshmi, P. Premchand, and A. Govardhan . “An Approach for Tongue Diagnosing with
Sequential Image Processing Method’’, International Journal of Computer Theory and Engineering,
Vol. 4, No. 3, June 2012
[3].Jia Wu; Yonghong Zhang; Jing Bai, "Tongue Area Extraction in Tongue Diagnosis of Traditional
Chinese Medicine," in Engineering in Medicine and Biology Society, 2005. IEEE-EMBS 2005. 27th
Annual International Conference of the, vol., no., pp.4955-4957, 17-18 Jan. 2006
10 A.Selvarani and G.R.Suresh
[4]Li.C.H, and P.C. Yuen, “Tongue image matching using color content, “Pattern Recognition’’, vol.
35, pp. 407-419, 2002
[5] Shen.L.S, Wei.B.G, “Image analysis for tongue characterization”, Acta Electronica Sinica, 2001,
pp.1762-1765
[6] Xingzheng Wang; Zhang, B.; Zhimin Yang; Haoqian Wang; Zhang, D., "Statistical Analysis of
Tongue Images for Feature Extraction and Diagnostics," in Image Processing, IEEE Transactions on,
vol.22, no.12, pp.5336-53
