Applied Soft Computing 36 (2015) 210217

Contents lists available at ScienceDirect

Applied Soft Computing

journal homepage: www.elsevier.com/locate/asoc

A novel 3D skeleton algorithm based on neutrosophic cost function

Yanhui Guo a, , Abdulkadir Sengur b
a
Department of Computer Science, University of Illinois at Springeld, Springeld, IL 62703, United States
b
Department of Electric and Electronics Engineering, Firat University, Elazig, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: A skeleton provides a synthetic and thin representation of three dimensional objects, and is useful for
Received 21 October 2014 shape description and recognition. In this paper, a novel 3D skeleton algorithm is proposed based on
Received in revised form 23 July 2015 neutrosophic cost function. Firstly, the distance transform is used to a 3D volume, and the distance
Accepted 24 July 2015
matrix is obtained for each voxel in the volume. The ridge points are identied based on their distance
Available online 31 July 2015
transform values and are used as the candidates for the skeleton. Then, a novel cost function, namely
neutrosophic cost function (NCF) is proposed based on neutrosophic set, and is utilized to dene the cost
Keywords:
between each ridge points. Finally, a shortest path nding algorithm is used to identify the optimum
Skeleton
3D volume
path in the 3D volume with least cost, in which the costs of paths are calculated using the new dened
Neutrosophic set NCF. The optimum path is treated as the skeleton of the 3D volume. A variety of experiments have been
Distance transform conducted on different 3D volume. The experimental results demonstrate the better performance of the
Path nding algorithm proposed method. It can identify the skeleton for different volumes with high accuracy. In addition, the
proposed method is robust to the noise on the volume. This advantage will lead it to wide application in
the skeleton detection applications in the real world.
2015 Elsevier B.V. All rights reserved.

1. Introduction one-dimensional object (a set of curves embedded in 3D) which

captures a 3D objects main symmetry axes, but does not contain
A skeleton is dened as the locus of the centers of maximally detailed information about the object boundary. Skeletons provide
inscribed balls [1]. It can serve as a representation of the shape, and a compact, efcient, and simple description for tubular structures
it contains all the information necessary to reconstruct the shape. such as blood vessels [6], neurons, or elongated organs such as the
Skeletons have been used in several applications in computer human colon [7,8].
vision, image analysis, and digital image processing, including opti- A number of skeleton algorithms have been proposed for the
cal character recognition, ngerprint recognition, visual inspection, 3D voxel volume, and most of them can be categorized into three
pattern recognition, binary image compression, and protein folding classes [4]: topological thinning, distance transform based, and
[2]. polygon based methods. We focus here on voxel based volume in
In a three dimensional (3D) volume, a skeleton generally con- this paper. Therefore, we only discuss the rst two classes.
sists of both curves and surfaces. It provides a synthetic and thin The topological thinning methods extract the skeleton [9,10]
representation of objects. The 3D skeleton algorithm can be applied through iteratively removing its simple points from the bound-
into computer vision, 3D object feature extraction, object recog- ary of a voxel set, which does not alter the object topology [4].
nition and tracking, shape recognition and retrieval, 3D surface The skeleton is extracted using the distance eld [11] or a rened
reconstruction, automatic medical navigation and visualization, geodesic eld [12] in the 3D volume. Thinning method produces a
and geometric modeling [3,4]. Especially, it plays an important role connected and one-voxel thick structure [9,13]. However, it might
in computer assisted navigation and surgery. lead to many undesired small branches [4].
A 3D skeleton is a collection of points, lines, and surfaces [4,5]. The distance based methods compute the volumes distance
A skeleton is closely related to the centerline. A skeleton is a transform value and dene the skeleton as a set of the local maxima
on the distance transform values. This denition produces sur-
face skeletons, not a one dimensional centerline. The skeletons are
Corresponding author. Tel.: +1 4352275882. usually disconnected, not guaranteed to be one voxel thick, and sen-
E-mail addresses: [email protected] (Y. Guo), [email protected] (A. Sengur). sitive to boundary noise [4]. Zhou et al. [14] connect the directly

http://dx.doi.org/10.1016/j.asoc.2015.07.025
1568-4946/ 2015 Elsevier B.V. All rights reserved.
 Y. Guo, A. Sengur / Applied Soft Computing 36 (2015) 210217 211

extracted local maxima by local maximum paths. However, this 2.2. Ridge point
method is still sensitive to the noise. Moreover, the local maximum
extraction might produce surface skeletons and not centerlines. According to the denition of skeleton, the points on the
Several distance transform based methods were proposed to skeleton usually are inside the volume and have bigger distance
trade speed for precision such as Chamfer-based [15], level-set transform values than those of their neighbors, which are named
based [16], and true Euclidean [17]. Zhou et al. used a second dis- as ridge points in the distance map. A ridge point can represent one
tance eld to encode the distance to a given seed voxel set [8]. convex part on the volume, and can be used to represent the voxel
All voxels with the same second distance eld value and the one on the skeleton.
having maximal distance transform value are added to the cen-
terline. Bitter et al. [18] computed the gradient on the distance Denition 1. Ridge point. Let p be a voxel, and DM(p) is pixel ps
transform and found voxels on and close to the centerline by ana- distance transform value. Q is ps neighbors set in 3D. p is a ridge
lyzing the gradients local variations, followed by the reconnection point if it satises the equation:
of these structures. The nal centerline is computed by applying
various cost-based path tracing methods on the connected set of DM(p) DM(q), q Q (4)
extracted voxels. However, this algorithm does not handle objects
with branches or holes. It is also not very clear how sensitive the The neighborhood can be 26-voxel in 3D volume. A ridge point
algorithm is to the distance transform gradient value. is dened as the pixel which has biggest value among its neighbor-
In the existing methods, some of them are time-consuming and hood. The ridge points on the distance transform have the largest
sensitive to noise, and skeletons obtained by them are not sin- distance among their neighborhood, which guarantee these pixels
gle voxel. In this paper, to overcome the above disadvantages, we are the in the central part of the object. These ridge pixels can be
propose a 3D skeleton algorithm based on neutrosophic set. The used as candidate points on the skeleton and nd the one pixel line
skeleton is single voxel curve and immune to noise. At rst, a dis- as the skeleton of the object which is in the most middle of the
tance map is obtained using a 3D distance transform, and a ridge object.
point set is dened in the distance map. Then a cost function of each Denition 2. Ridge point set RPS is a set, in which each element
ridge point is dened based on the true and indeterminacy value is a ridge point.
in neutrosphic set. Finally, the skeleton is found as the lowest cost
path between different ridge points. RPS = {p|RPS, DM(p) DM(q), q Q } (5)
The paper is organized as follows. Section 2 describes the pro-
posed method, Section 3 discusses the experimental results and The ridge point set can be seen as the candidate of the skeleton.
comparisons, and the conclusions are drawn in Section 4. However, the ridge points are not connected, and easily affected
by the noisy volumes or/and small convex objects, which make the
2. Proposed method skeleton not smooth and having a variety of small burrs.

A novel skeleton algorithm is proposed based on neutrosophic

set for 3D images. At rst, the distance map of the 3D volume 2.3. Cost function
is generated using a distance transform. Then ridge point set is
dened and selected according the distance map value as the skele- Neutrosophic theory is as a new branch of philosophy dealing
ton candidate pixels. The neutrosophic set is used to dene the with the origin, nature and scope of neutralities [21]. In neutroso-
cost function for the path nding algorithm, and the denition can phy theory, every event has a truth degree, a falsity degree and an
reduce the noise effect for the indeterminacy is considered. Finally, indeterminacy degree [22]. Thus, neutrosophic set provides a pow-
the path nding algorithm is utilized to nd the path with minimum erful tool to deal with the indeterminacy for different applications.
cost, which is regarded as the skeleton of the 3D volume. To make the skeleton more smooth and robust to noise, we
employ neutrosophic set to dene the cost function for path nding
algorithm. The indeterminacy of the cost function will be elimi-
2.1. Distance transform
nated using the new dened function.
At rst, the truth degree (T) and indeterminacy degree (I) for
A distance transform (DT) can be used as a representation of
each voxel were dened using the distance map value.
a binary image, and it labels each pixel in the image with the dis-
tance to the closest boundary [19]. DT is a fundamental geometrical Denition 3. Neutrosophic set value Ns is used to measure the
operator in computer vision and graphics, shape analysis, pattern true and indeterminacy membership value:
recognition, and computational geometry [20].
Ns(p) = (1 T (p)) I(p) (6)
DT (pi ) = min{d(pi , pj )} pj B (1)
 DM(p)
d(pm , pn )=
 (xm xn )2 + (ym yn )2 + (zm zn )2 (2) T (p) = NDM(p) = (7)
max(DM)
where pi is a object voxel, and pj is a voxel in the background set B.
GDM(p)
d(pm , pn ) is a distance between any two voxels pm and pn . I(p) = NGDM(p) = (8)
In the proposed method, the Euclidean distance metric is used max(GDM(p))
to measure the pixels distance, which is the adequate model to
where DM and GDM are the value on the distance transform map
numerous geometrical objects. A distance map DM is obtained
and gradient value on distance transform map, and NDM and GDM
when distance transform is employed on a volume V.
are their normalized results, respectively.
DM(pi ) = DT (pi ) (3)
Denition 4. Cost function C between two neighboring voxels p
where DT( ) is a distance transform function. V is the original volume and q, and its cost function value C(p,q) is computed as:
of the 3D image, pi is a object voxel in V, and DM is the distance map
of V. C(p, q) = Ns(p) + Ns(q) (9)
212 Y. Guo, A. Sengur / Applied Soft Computing 36 (2015) 210217

Fig. 1. A pig and its skeleton result at different views.

Fig. 2. A dog and its skeleton result at different views.

 Y. Guo, A. Sengur / Applied Soft Computing 36 (2015) 210217 213

Fig. 3. A horse and its skeleton result at different views.

Denition 5. Cost function of a path C(P). Let P = {p1 , p2 , . . . pn } 8. Merge the close branches and prune the small branches;
be a connected path. The path P has n points, and its cost function 9. Pick the paths as the volumes skeleton SkV.
value is:

n 3. Experimental results and discussion
C(P) = Ns(pi ) (10)
i=1
A variety of 3D volume are obtained from the 3D models from
the websites1,2 , and used to test the proposed methods perfor-
The denition of cost function can guarantee the skeleton has mance. As mentioned in literature [24], it is hard to compare the
the lowest cost. quality of the skeletons generated by different methods because
there has been no numerical criterion to evaluate the difference
2.4. 3D skeleton algorithm between the skeleton and the original 3D volume. We demonstrate
the proposed methods performance via the visual evaluation way
After the ridge point set is dened and the candidate points on as in the literature.
the skeleton are determined, a path nding algorithm is utilized Five examples are employed to show the skeleton results in
to nd the paths between ridge points, which is regarded as the Figs. 15. To display the details clearly, we show 3D objects and
skeleton of the volume. A graph algorithm, Dijkstras algorithm [23] their skeletons at different orientation angles. In Fig. 1, a pig is
is used to nd the lowest cost path between different ridge points. shown in (a) and (c) at different angles in 3D view. Fig. 1(b) and
The steps to determine the skeleton can be briey described as (d) are the skeleton at the corresponding angles, respectively. In
follows: the skeleton result, not only the main body of the big, such as body
and legs, was described by the skeleton, but also some details, such
Algorithm: 3D skeleton algorithm based on neutrosophic set as ears, feet and tail, are preserved and extracted properly.
Input: 3D volume V. Fig. 2 shows another animal model, a 3D dog object and its
Output: 3D skeleton line SkV. skeleton result by the proposed method. The results displayed at
Steps: different angles show that the dogs skeleton very well, including
1. Calculate the distance transform on the volume V to obtain the four legs and long tail. The dogs skeleton can represent the whole
distance map DM(V) and true membership T(V) in NS; 3D dogs shape and structure correctly.
2. Compute the gradient value on DM(V) to obtain the indetermi- Fig. 3 shows a 3D horse volume and its skeleton result. The
nacy value I(V); horses body and legs are also described by the skeleton very
3. Calculate the neutrosophic value Ns(V); smooth and precise. In addition, its two small ears are extracted
4. Construct the ridge point set RPS; with very high accuracy.
5. Calculate the cost values for each point in RPS. The cost value of
the key point RPi is dened as C(i) = Ns(i);
6. Select one point from RPS randomly as the start point SP; 1
http://www.cc.gatech.edu/projects/large models/.
7. Find the lowest cost paths from SP to all points in RPS; 2
http://liris.cnrs.fr/meshbenchmark/.
214 Y. Guo, A. Sengur / Applied Soft Computing 36 (2015) 210217

Fig. 4. A hand and its skeleton result at different view.

A hand model is shown in Fig. 4. The skeletons of ve ngers A cow model is shown in Fig. 5 to demonstrate the proposed
are extracted with very high accuracy. The hands shape can be methods advantages and limitation. The skeletons of cow body,
presented clear, which is useful to the 3D shape recognition and four legs are extracted with very high accuracy. Especially, two
gesture recognition. ears and mouth are also extracted correctly. However, the tail part

Fig. 5. A cow and its skeleton result at different view.

 Y. Guo, A. Sengur / Applied Soft Computing 36 (2015) 210217 215

Fig. 6. A shark and its skeleton result at different view.

does not have the correct skeleton, especially for the connection Kerschnitzki et al. proposed a novel 3D skeleton algorithm,
between tail and body. It is due to our proposed method try to named homotopic thinning algorithm, based on distance transform
nd the shortest path between the body and tail parts, and connect and neighborhood connection to describe the structure of bone and
them directly as the skeleton. When there is a hole in the volume, vessels [25]. The homotopic thinning algorithm achieved a good
the proposed method will fail to nd the correct skeleton. result in 3D volume and had been applied into medical application
Figs. 6 and 7 show a sharks and triceratopss skeletons. From the [25]. To demonstrate the performance on 3D skeleton, we com-
results, we can see clearly the results have smooth main skeleton pare the proposed method with the homotopic thinning algorithm.
and less small branch. The skeleton on ns of shark and legs of Figs. 810 show the comparison results, in which the left columns
triceratops are extracted correctly. are the homotopic thinning algorithms results and the proposed

Fig. 7. A triceratops and its skeleton result at different view.

216 Y. Guo, A. Sengur / Applied Soft Computing 36 (2015) 210217

Fig. 8. Dogs skeleton result using different algorithm.

Fig. 9. Pigs skeleton result using different algorithm.

Fig. 10. Hands skeleton result using different algorithm.

methods results are displayed in the right ones. From the com- foundation for wide applications in 3D volume presentation and
parison results, we can see the proposed method achieved better shape recognition.
results with smooth skeleton and keeping objects shape, while the
homotopic thinning methods skeletons have small branches and
References
are burry, especially on the dog and pigs body, and palm regions.
[1] A. Lieutier, Any open bounded subset of Rn has the same homotopy type as its
medial axis, Comput.-Aided Des. 36 (11) (2004) 10291046.
4. Conclusion [2] S. Abeysinghe, T. Ju, M.L. Baker, W. Chiu, Shape modeling and matching in
identifying 3D protein structures, Comput.-Aided Des. 40 (6) (2008) 708720.
In this paper, a 3D skeleton algorithm has been developed to [3] N.D. Cornea, D. Silver, P. Min, Curve-skeleton properties, applications, and algo-
rithms, IEEE Trans. Vis. Comput. Graph. 13 (3) (2007) 530548.
identify the skeleton on 3D volume. It is insensitive to the size
[4] A. Telea, A. Vilanova, A robust level-set algorithm for centerline extraction, pp.
variations and noise of the volumes due to the employment of 185194.
neutrosophic set for indeterminacy processing. Through the exper- [5] R. Kimmel, D. Shaked, N. Kiryati, A.M. Bruckstein, Skeletonization vis distance
iments, we observe that the skeleton can be distinguished correctly maps and level sets, Comput. Vis. Image Underst. 62 (3) (1995) 382391.
[6] D. Lesage, E.D. Angelini, I. Bloch, G. Funka-Lea, A review of 3D vessel lumen seg-
and accurately. The proposed method achieved better performance mentation techniques: models, features and extraction schemes, Med. Image
with high accuracy. This result is encouraging, and lays a solid Anal. 13 (6) (2009) 819845.
 Y. Guo, A. Sengur / Applied Soft Computing 36 (2015) 210217 217

[7] A. Vilanova, A. Konig, E. Groller, VirEn: a virtual endoscopy system, Mach. Graph. Signal Processing, Computational Imaging and Vision, Springer, US, 2000, pp.
Vis. 8 (3) (1999) 469487. 331340.
[8] Y. Zhou, A. Kaufman, A.W. Toga, Three-dimensional skeleton and centerline [18] I. Bitter, M. Sato, M. Bender, K.T. McDonnell, A. Kaufman, M. Wan, CEASAR: a
generation based on an approximate minimum distance eld, Vis. Comput. 14 smooth, accurate and robust centerline extraction algorithm, pp. 4552.
(7) (1998) 303314. [19] A. Rosenfeld, J.L. Pfaltz, Distance functions on digital pictures, Pattern Recognit.
[9] N. Gagvani, D. Silver, Parameter-controlled volume thinning, Graph. Models 1 (1) (1968) 3361.
Image Process. 61 (3) (1999) 149164. [20] R. Fabbri, L.D.F. Costa, J.C. Torelli, O.M. Bruno, 2D Euclidean distance transform
[10] S.M. Yoon, C. Malerczyk, H. Graf, 3D skeleton extraction from volume data based algorithms: a comparative survey, ACM Comput. Surv. (CSUR) 40 (1) (2008)
on normalized gradient vector ow, pp. 177182. 2.
[11] F.-C. Wu, W.-C. Ma, P.-C. Liou, R.-H. Laing, M. Ouhyoung, Skeleton extraction of [21] F. Smarandache, A Unifying Field in Logics: Neutrosophic Logic. Neutrosophy,
3d objects with visible repulsive force, pp. 124131. Neutrosophic Set, Neutrosophic Probability: Neutrsophic Logic. Neutrosophy,
[12] T.K. Dey, J. Sun, Dening and Computing Curve-skeletons with Medial Geodesic Neutrosophic Set, Neutrosophic Probability: Innite Study, 2005.
Function, pp. 143152. [22] W.V. Kandasamy, F. Smarandache, Basic neutrosophic algebraic structures
[13] A. Manzanera, T.M. Bernard, F. Pretux, B. Longuet, Medial faces from a concise and their application to fuzzy and neutrosophic models: Innite Study,
3D thinning algorithm, pp. 337343. 2004.
[14] Y. Zhou, A.W. Toga, Efcient skeletonization of volumetric objects, IEEE Trans. [23] E.W. Dijkstra, A note on two problems in connexion with graphs, Numer. Math.
Vis. Comput. Graph. 5 (3) (1999) 196209. 1 (1) (1959) 269271.
[15] A. Telea, J.J. Van Wijk, An augmented fast marching method for computing [24] D. Khromov, L. Mestetskiy, 3D curve-skeletons extraction and evaluation, in:
skeletons and centerlines, pp. 251260. 21st International Conference on Computer Graphics and Vision, Moscow, Rus-
[16] M. Rumpf, A. Telea, A Continuous Skeletonization Method Based on Level Sets, sia, 2011.
pp. 151-157. [25] M. Kerschnitzki, P. Kollmannsberger, M. Burghammer, G.N. Duda, R.
[17] A. Meijster, J.B.T.M. Roerdink, W.H. Hesselink, A general algorithm for com- Weinkamer, W. Wagermaier, P. Fratzl, Architecture of the osteocyte net-
puting distance transforms in linear time, in: J. Goutsias, L. Vincent, D. work correlates with bone material quality, J. Bone Miner. Res. 28 (8) (2013)
Bloomberg (Eds.), Mathematical Morphology and its Applications to Image and 18371845.