International Journal of Electrical and Electronics Research ISSN 2348-6988 (online)

Vol. 2, Issue 3, pp: (51-57), Month: July - September 2014, Available at: www.researchpublish.com

Page | 51

Research Publish Journals

CLG for Automatic Image Segmentation

Christo Ananth1, S.Santhana Priya2, S.Manisha3,

T.Ezhil Jothi4, M.S.Ramasubhaeswari5

1 Assistant Professor/ECE, Francis Xavier Engineering College, Tirunelveli

2,3,4,5 U.G.Scholar/ECE, Francis Xavier Engineering College, Tirunelveli, Tamil Nadu, India

Abstract: This paper proposes an automatic segmentation method which effectively combines Active Contour

Model, Live Wire method and Graph Cut approach (CLG). The aim of Live wire method is to provide control to

the user on segmentation process during execution. Active Contour Model provides a statistical model of object

shape and appearance to a new image which are built during a training phase. In the graph cut technique, each

pixel is represented as a node and the distance between those nodes is represented as edges. In graph theory, a cut

is a partition of the nodes that divides the graph into two disjoint subsets. For initialization, a pseudo strategy is

employed and the organs are segmented slice by slice through the OACAM (Oriented Active Contour Appearance

Model). Initialization provides rough object localization and shape constraints which produce refined delineation.

This method is tested with different set of images including CT and MR images (3D image) and produced perfect

segmentation results.

Keywords: Oriented Active Contour Appearance Model, Active Contour Model, Computed Tomography, Live

Wire method, Graph Cut approach.

I. INTRODUCTION

CIRHOSSIS, lymphoma, pancreatitis, Hodgkin’s disease, and renal cell carcinoma, are just a few of the many diseases

that can be diagnosed using CT scans of the abdomen. The development of computer-aided diagnosis systems would

allow anatomical knowledge coupled with image processing techniques to improve healthcare. Knowledge of anatomy is

only of utility if that knowledge is applied to the removal of disease and to the preservation of health. Computer-based

systems for the analysis of CT images have many advantages over human interpreters, such as speed, large knowledge

base for diagnostic information, and non-sensitivity to fatigue. Organ segmentation is often the first step in computer-

aided diagnosis. Segmentation of abdominal organs, such as the liver, kidneys, and spleen, from CT scan imagery has

been attracting a fair amount of research recently. Segmentation of abdominal organs presents many challenges. Many

artifacts can arise in CT scans, among these are beam-hardening artifacts, which are noticeable as focal areas of low

attenuation adjacent to bones; partial-volume artifacts, resulting from spatial averaging of disparate tissues in close

proximity and resulting in blurred edges; and streak artifacts, the result of peristalsis, respiratory, cardiac, and patient

motion. Also, different organs and tissues have very similar gray levels, which consign thresholding to limited utility.

In [1], the authors devise a graph cut algorithm for interactive segmentation which incorporates shape priors. Interactive

or semi-automatic segmentation is a useful alternative to pure automatic segmentation in many applications the graph cuts

approach guarantees a global optimum. The related work falls into two categories: segmentation using shape priors, and

globally optimal methods for segmentation. The shape priors are embedded into the weights on the edges in the graph, by

using a level-set formulation. Transformations of the shape template are also taken into account. The main direction for

future research is to examine whether more complex transformations of the template can be easily incorporated into the

scheme.

In [2], the proposed algorithm utilizes the shape model of the target organ to gain robustness in the case where the

objective organ is surrounded by other organs or tissue with the similar intensity profile. The algorithm labels the image

based on the graph-cuts technique and incorporates the shape prior using a technique based on level-sets. The method

requires proper registration of the shape template for an accurate segmentation, so propose a unified registration-

segmentation framework to solve this problem. In order to reduce the computational cost of the minimization step, the

International Journal of Electrical and Electronics Research ISSN 2348-6988 (online)

Vol. 2, Issue 3, pp: (51-57), Month: July - September 2014, Available at: www.researchpublish.com

Page | 52

Research Publish Journals

proposed segmentation algorithm operates on homogeneous regions instead of voxels. The algorithm, operating on a

properly registered template, captures the boundary of the object, even if it is diffuse or weak. The registration algorithm

uses the segmentation energy as a measure of how well the template is fit to the input image, and minimizes that energy

over the space of transformations. In this way, the proposed algorithm registers the template and segments the image

simultaneously. In [3], In this paper, a probabilistic method for segmenting instances of a particular object category within

an image is explained. This approach overcomes the deficiencies of previous segmentation techniques based on traditional

grid conditional random fields (CRF), namely that 1) they require the user to provide seed pixels for the foreground and

the background and 2) they provide a poor prior for specific shapes due to the small neighbourhood size of grid CRF.

Automatically obtain the pose of the object in a given image instead of relying on manual interaction. This “objcut”

method include: 1) efficient algorithms for sampling the object category models and 2) the observation that a sampling-

based approximation of the expected log-likelihood of the model can be increased by a single graph cut. This method

efficiently provides accurate segmentation which resembles the object. The accuracy of the segmentation can be attributed

to the novel probabilistic model.

II. MATERIALS AND METHODS

A. Image Segmentation Methods

Image-based methods perform segmentation based only on information available in the image; these include thresholding,

region growing, morphological operations, active contours, level sets, live wire (LW), watershed, fuzzy connectedness,

and graph cuts (GCs). These methods perform well on high-quality images. However, the results are not as good when the

image quality is inferior or boundary information is missing.

Model-based methods employ object population shape and appearance priors such as atlases, statistical active shape

models, deformable templates and statistical active appearance models (ACMs). When some object information is

missing, such gaps can be filled by drawing upon the prior information present in the model.

Hybrid methods that form a combination of two or more approaches are emerging as powerful segmentation tools. The

synergy that exists between these two approaches, i.e., purely image-based and model-based strategies is called as hybrid

methods. Hybrid approach can achieve a result much quicker with greater accuracy.

Most of the image-based, model-based and even hybrid segmentation techniques are often tailored for specific body

regions (brain, abdomen, etc.) and different image modalities (CT, MRI, etc.). However, it is desirable to generalize

image segmentation methodologies for any (or most) body regions and different image modalities and protocols.

Furthermore, it is desirable for an image segmentation algorithm not to heavily depend upon the characteristics of fixed

shape families and different image modalities. While perhaps some of the above techniques can be generalized in this

spirit, few methods have demonstrated to work in this general setting.

B. Active Contour Model

An Active Contour Model (ACM) is a computer vision algorithm for matching a statistical model of object shape and

appearance to a new image which are built during a training phase. A set of images, together with coordinates of

landmarks that appear in all of the images, is provided to the training supervisor. The approach is widely used for

matching and tracking faces and for medical image interpretation. The algorithm uses the difference between the current

estimate of appearance and the target image to drive an optimization process. By taking advantage of the least squares

techniques, it can match to new images very swiftly. It is related to the active shape model (ASM). One disadvantage of

ASM is that it only uses shape constraints, and does not take advantage of all the available information – the texture

across the target object. This can be modeled using an ACM

C. Live Wire Method

This method (LW) determines object boundary information from orthogonal slices of a volume segmented by a user. This

technique allows the user to segment a minimal number of slices, reducing the total segmentation time. The user-steered

2-D segmentation method allows the user to select three seed points. After selection, Live Wire is used to generate an

initial frame and then “spokes.” After the spokes are found, a gap filling algorithm is used to close the space between

spokes. In Live Wire method, the user clicks on an edge of the object with the mouse, defining a “seed point”. The user

International Journal of Electrical and Electronics Research ISSN 2348-6988 (online)

Vol. 2, Issue 3, pp: (51-57), Month: July - September 2014, Available at: www.researchpublish.com

Page | 53

Research Publish Journals

then moves the cursor to some other portion of the object’s edge. The pixel that lies under the cursor is called the “free

point”. As the free point moves, a wire connecting the seed point and the free point automatically snaps to the edge.

A drawback of this method is that the speed of optimal path computation depends on image size. On modestly powered

computers, for images of even modest size, some sluggishness appears in user interaction, which reduces the overall

segmentation efficiency. In this work, this problem is solved by exploiting some known properties of graphs to avoid

unnecessary minimum-cost path computation during segmentation.

D. Graph Cut Approach

Graph cuts (GC) can be employed to efficiently solve a wide variety of low-level computer vision problems, such as

image smoothing, the stereo correspondence problem, and many other computer vision problems that can be formulated

in terms of energy minimization. Such energy minimization problems can be reduced to instances of the maximum flow

problem in a graph .Under most formulations of such problems in computer vision, the minimum energy solution

corresponds to the maximum a posteriori estimate of a solution. Although many computer vision algorithms involve

cutting a graph (e.g., normalized cuts), the term "graph cuts" is applied specifically to those models which employ a max-

flow/min-cut optimization."Binary" problems (such as denoising a binary image) can be solved exactly using this

approach; problems where pixels can be labelled with more than two different labels (such as stereo correspondence, or

denoising of a gray scale image) cannot be solved exactly, but solutions produced are usually near the global optimum.

The basic cuts in the graph theory are minimum cut and maximum cut. In the minimum cut technique the size of the cut

is not larger than the size of the any other cut. In the minimum cut technique the size of the cut is not larger than the size

of the any other cut. A cut is maximum if the size of the cut is not smaller than the size of any other cut in the image. In

the minimum curt technique we need to segment each and every pixel of an image. That means in this technique we need

to cut each pixel in an image even if those pixels are similar with respect to color or intensity or texture. So in this

minimum cut technique we couldn’t get a better segmented image when compared to the other techniques so here we use

the normalized cut method to segment the image.

E. Limitations of ACM, LW and GC Approaches

The main limitation of LW stems from the recognition process, wherein the anchor points are to be selected on the

boundary by a human operator. The specific shape and appearance information on the object in a given image is difficult

to account for in ACM. GC methods have the ability to compute globally optimal solutions (in the two-label case) and can

enforce piecewise smoothness. However, they are interactive methods, requiring labelling of the source and sink seeds by

a human operator.

III. PROPOSED METHODOLOGY

A. Training Phase

Fig.1.Training Phase

Fig.1. shows the Training Phase model. Before building the model, the top and bottom slices of each organ are first

manually identified. Then, linear interpolation is applied to generate the same number of slices for the organ in every

training image. This is for establishing anatomical correspondences. 2-D OACAM models are then constructed for each

slice level from the images in the training set. The LW cost function and GC parameters are also estimated in this stage.

Interpolation is a method of constructing new data points within the range of a discrete set of known data points. In

engineering and science, one often has a number of data points, obtained by sampling or experimentation, which represent

the values of a function for a limited number of values of the independent variable. It is often required to interpolate (i.e.

estimate) the value of that function for an intermediate value of the independent variable. This may be achieved by curve

fitting or regression analysis. Suppose the formula for some given function is known, but too complex to evaluate

Training Images

Linear

Interpolation

Land Marking

Construction of ACM,

Live wire and Graph Cut

International Journal of Electrical and Electronics Research ISSN 2348-6988 (online)

Vol. 2, Issue 3, pp: (51-57), Month: July - September 2014, Available at: www.researchpublish.com

Page | 54

Research Publish Journals

efficiently. A few known data points from the original function can be used to create an interpolation based on a simpler

function. Of course, when a simple function is used to estimate data points from the original, interpolation errors are

usually present; however, depending on the problem domain and the interpolation method used, the gain in simplicity may

be of greater value than the resultant loss in accuracy. The interpolation methods are: Piecewise constant interpolation,

Linear interpolation, Polynomial interpolation and Spline interpolation. Linear interpolation is a method of curve fitting

using linear polynomials.. In this project, linear interpolation is used to generate the same number of slices for the organ

in every training image.

Although semiautomatic or automatic methods are also available for annotating organs because of its simplicity,

generality, and efficiency, manual land marking is still in use in clinical research. Therefore, manual land marking is used

to annotate organs’ shape. In manual land marking, trained operators identify prominent landmarks on each shape visually

on displayed slices. We assessed a semiautomatic land marking method, which is called equal-space land marking [6] to

show that there is a strong correlation between the shapes encoded by the manual and semi automated land marking

methods. Once the landmarks are specified, the standard ACM method is used for constructing the model. The model

includes both shape and texture information. Suppose represents the ACM model for slice level and the number of

slice levels is , then the complete model can be represented as = ( , , ,….. ). Similar to the oriented

active shape model method [16], an oriented boundary cost function is devised for each organ included in the model as

per the LW method [9]. Following the original terminology and notation in [9], let define a boundary element (bel) as an

oriented edge between two pixels with values 1 and 0. In our particular case, the cost associated with bel is a linear

combination of the costs assigned to its features

nf

i=1

(1)

where nf is the number of features, is a positive constant, and is the function to convert feature values at to

cost values . In LW, , may represent features such as intensity on the immediate interior of the boundary,

intensity on the immediate exterior of the boundary, and gradient magnitude at the center of the bel.

A. Segmentation Phase

Fig.2.Segmentation Phase

Fig.2. shows the Segmentation Phase model. The aim of slice localization in our approach is to locate the top and bottom

slices of the organ. Since the model is already trained for each organ slice, we can use the model for slice localization.

The proposed method is based on the similarity of the slice to the OACAM model. For the localization of a top slice in a

given image, the top slice model is applied to each slice in the image using the recognition method which is explained

Input Image

Localization

Object Recognition

Shape Refinement

Delineation

Output Segmentation

Results

International Journal of Electrical and Electronics Research ISSN 2348-6988 (online)

Vol. 2, Issue 3, pp: (51-57), Month: July - September 2014, Available at: www.researchpublish.com

Page | 55

Research Publish Journals

detail in below Section. Then, the slice corresponding to maximal similarity (minimal distance) is taken as the top slice of

the organ.

The initialization step plays a key role in our method, which provides the shape constraints to the later GC delineation

step and makes it fully automatic. The proposed initialization method includes three main steps. First, a slice localization

method is applied to detect the top and bottom slices of the organ. Next a linear interpolation is applied to generate the

same number of slices for the given image of a subject, as in the model. Then, the organ is recognized slice by slice via

the OACAM method. The proposed object recognition method is based on the ACM. The conventional ACM matching

method for object recognition is based on the RMS difference between the appearance model instance and the target

image. Such a strategy is better suited for matching appearances than for the detailed segmentation of target images.

The conventional ACM searching method is performed once to obtain a rough placement of the model. Then, the

following method is applied to refine the shape model in the ACM. The shape is extracted from the shape model of the

ACM, and then the landmarks are updated based on LW using only the shape model and the pose parameters.

Subsequently, the refined shape model is transformed back into the ACM. At the same time, ACM refinement is applied

to the image yielding its own set of coefficients for shape and pose. The purpose of this step is to finally precisely

delineate the shapes recognized in the previous step. An iterative algorithm is proposed by combining GC and OACAM

method for the organ’s delineation. It effectively integrates the shape information with the globally optimal delineation

capability of the GC method.

GC segmentation can be formulated as an energy minimization problem such that, for a set of pixels p and a set of labels,

the goal is to find a labelling that minimizes the energy function En as follows:

En = + (2)

where is the set of pixels in the neighbourhood of is the cost of assigning label,

the cost of assigning labels .

The proposed shape-integrated energy function is defined as follows:

En = + (3)

where, , and are the weights for the data term, shape term , and the boundary term, respectively. The

minimization of the equation (3) can be solved by the GC method.

IV. RESULTS AND DISCUSSION

The training images are processed with live wire method based on specified landmarks. The landmarks points are

specified through “red star” in each images are shown in below. There are 33 landmarks have been used in each training

image for constructing Liver object segmentation. The live wire method can detect the strong edges between the starting

points to next point and so on.

Fig.3.Shape Model of Training Image.

International Journal of Electrical and Electronics Research ISSN 2348-6988 (online)

Vol. 2, Issue 3, pp: (51-57), Month: July - September 2014, Available at: www.researchpublish.com

Page | 56

Research Publish Journals

Fig.3. shows the shape model of the training image. It can find the best shape of the specified object of the each training

images. Next, the object is applying to the appearances model for constructing texture information by using the parameter

of PCA with the output of shape model of the training images.

Fig.4.PCA Output

Fig.4. shows the image which is applied for PCA to get the mean shape of the ACM model. The next model is combined

model of texture and shape outputs of the images. In this combined model can give a output of RMS difference between

the original texture and the combined model.

Fig.5. ACM Output

Fig.5. shows the RMs difference between the original texture and the combined model. This is the ACM Output.

Fig.6. (a)Test Image (b) Selection of Contour Point (c) Contour Drawn of the Abdominal Output (d) Comparisons of

Segmented Liver Image With Similar Model Image (e) Final 2-D Segmentation.

International Journal of Electrical and Electronics Research ISSN 2348-6988 (online)

Vol. 2, Issue 3, pp: (51-57), Month: July - September 2014, Available at: www.researchpublish.com

Page | 57

Research Publish Journals

The best contour line is fixed manually to the best place of test image. This contour point is selected by using automated

Live Wire method. The finalized segmented output, which is the representation of the proposed algorithm for the 2D test

image is as shown in Fig.6.

V. CONCLUSION

This paper presented an automatic segmentation method which effectively combines Active Contour Model, Live Wire

method and Graph Cut approach (CLG). The aim of Live wire method is to provide control to the user on segmentation

process during execution. Active Contour Model provides a statistical model of object shape and appearance to a new

image which are built during a training phase. In the graph cut technique, each pixel is represented as a node and the

distance between those nodes is represented as edges. In graph theory, a cut is a partition of the nodes that divides the

graph into two disjoint subsets. For initialization, a pseudo strategy is employed and the organs are segmented slice by

slice through the OACAM (Oriented Active Contour Appearance Model). Initialization provides rough object localization

and shape constraints which produce refined delineation. This method is tested with different set of images including CT

and MR images especially 3D images and produced perfect segmentation results.

REFERENCES

[1] D. Freedman and T. Zhang, “Interactive graph cut based segmentation with shape priors,” in Proc. IEEE Comput.

Soc. Conf. Comput. Vis.Pattern Recogn., 2013, pp. 755–762.

[2] Ayvaci and D. Freedman, “Joint segmentation-registration of organs using geometric models,” in Proc. IEEE

Eng.Med. Biol. Soc., 2013,pp. 5251–5254.

[3] M. Kumar, P. Torr, and A. Zisserman, “ObjCut,” in Proc. IEEE Comput. Society Conference on Computer Vision

and Pattern Recognition, 2012, pp. 18–23.

[4] T. Kaneko, L. Gu, and H. Fujimoto, “Recognition of abdominal organs using 3D mathematical morphology,” Syst.

Comput. Jpn., vol. 33, no. 8, pp. 75–83, 2012.

[5] V. Kolmogorov and R. Zabih, “What energy functions can be minimized via graph cuts?,” IEEE Trans. Pattern

Anal. Mach. Intell., vol. 26, no. 2, pp. 147–159, Feb. 2011.

[6] J. Liu and J. Sun, “Parallel graph-cuts by adaptive bottom-up merging,” in Proc. IEEE Comput. Soc. Conf.

Comput. Vis. Pattern Recogn., 2010, pp. 2181–2188.

[7] J. M. Liu and J. K. Udupa, “Oriented active shape models,” IEEE Trans. Med. Imag., vol. 28, no. 4, pp. 571–584,

Apr. 2009.

[8] J. J. Malcolm, Y. Y. Rathi, and A. A. Tannenbaum, “Graph cut segmentation with nonlinear shape priors,” in

Proc. IEEE Int. Conf. Image Process., 2010, vol. IV, pp. 365–368.

[9] R. Malladi, J. A. Sethian, and B. C. Vemuri, “Shape modeling with front propagation: A level set approach,” IEEE

Trans. Pattern Anal.Mach. Intell., vol.17, no. 2, pp. 158–175, Feb. 2009..

[10] S. C. Mitchell, J. G. Bosch, B. P. F. Lelieveldt, R. J. van der Geest, J.H. C. Reiber, and M. Sonka, “3-D active

appearance models: Segmentation of cardiac MR and ultrasound images,” IEEE Trans. Med. Imag., vol. 21, no. 9,

pp. 1167–1178, Sep. 2008.

[11] S. C. Mitchell, B. P. F. Lelieveldt, R. J. van der Geest, H. G. Bosch, J. H. C. Reiber, and M. Sonka, “Multistage

hybrid active appearance model matching: Segmentation of left and right ventricles in cardiac MR images,” IEEE

Trans. Med. Imag., vol. 20, no. 5, pp. 415–423, May 2008.