Research Article A Novel Medical Freehand Sketch 3D Model ...

Research ArticleA Novel Medical Freehand Sketch 3D Model Retrieval Method byDimensionality Reduction and Feature Vector Transformation

Zhang Jing and Kang Bao Sheng

Northwestern University Xirsquoan Shanxi 710127 China

Correspondence should be addressed to Zhang Jing mywangwang123163com

Received 28 December 2015 Revised 14 April 2016 Accepted 26 April 2016

Academic Editor Syoji Kobashi

Copyright copy 2016 Z Jing and K B Sheng This is an open access article distributed under the Creative Commons AttributionLicense which permits unrestricted use distribution and reproduction in any medium provided the original work is properlycited

To assist physicians to quickly find the required 3Dmodel from themassmedicalmodel we propose a novel retrievalmethod calledDRFVT which combines the characteristics of dimensionality reduction (DR) and feature vector transformation (FVT) methodTheDRmethod reduces the dimensionality of feature vector only the top119872 low frequency Discrete Fourier Transform coefficientsare retained The FVT method does the transformation of the original feature vector and generates a new feature vector to solvethe problem of noise sensitivity The experiment results demonstrate that the DRFVTmethod achieves more effective and efficientretrieval results than other proposed methods

1 Introduction

Themedical 3Dmodel retrieval becomes a hot research topicdue to the rapid development in clinic and teaching Themedical 3D model can not only show doctors the anatomicalstructure of a particular part inside the human body butalso reveal the function of human organs to a certain degree[1] The current diagnostic techniques especially in thecases which cannot immediately get the diagnosis need theauxiliary diagnosis by previous similar cases At this time 3Dmodel retrieval technology is applied to obtain the similar3D model from the database In the face of vast numbersof medical 3D models how to find the required 3D modelrapidly and extract the valuable information accurately hasbecome an urgent issue [2]

The most popular way for retrieving 3D models isexample-based paradigm [3] where the user provides anexisting 3D model as query input and the retrieval methodcan return similar 3D models from the database Howeverit is difficult for a user to have an appropriate example 3Dmodel at hand An alternative way is to use a 2D sketch as aquery where users can describe a target 3D model by quicklydrawing it But a 2D sketch is merely a coarse and simplerepresentation which only contains partial information of an

original 3D model Hence it is more challenging to realizefreehand retrieval [4] than example-based retrieval

For the freehand sketch 3D model retrieval how to createefficient feature descriptors [5] is the most important partSeveral freehand sketch 3D model retrieval methods havebeen proposed recently Funkhouser et al [6] Chen et al [7]and Li et al [8] extracted feature descriptors from the outlineto describe the query sketch and 2D views of 3Dmodels suchas the spherical harmonics descriptor light field descriptorand shape context features RecentlyWang et al [9] provideda sketch-based retrieval approach by utilizing both globalfeature and local feature In the practical applications we findthat these features are not effective and efficient in evaluatingthe relevance between the query sketch and 3D modelsThere exist two main disadvantages for the current freehandsketch 3D model retrieval methods Firstly the traditionalFourier Transform [10] only considers describing the modeland the argument of the coefficient Compactness [11] andsimplicity of feature descriptors are necessary for minimizingthe storage overhead and the computing time Therefore itis necessary to describe the outline by a limited number ofcoefficients in the frequency domain Secondly a freehandsketch is a simple line drawing which has a high level ofabstraction inherent ambiguity and unavoidable noise [12]

Hindawi Publishing CorporationComputational and Mathematical Methods in MedicineVolume 2016 Article ID 4738391 11 pageshttpdxdoiorg10115520164738391

2 Computational and Mathematical Methods in Medicine

2D freehand sketch input

Start

Medical3D model set

2D projectionview of 3D model

Feature vectorextraction

Compare similarities 2D projection viewfeature vector set

Results of retrieval

End

2D outline representation

DR method

FVT method

2D sketchfeature vector set

Figure 1 The framework of the medical freehand sketch 3D model retrieval method

For effective retrieval the feature descriptors should be robustto noise and invariant to transformations

To tackle these problems this paper proposes a novelfreehand sketch 3D medical model retrieval method whichcombines the characteristics of dimensionality reduction(DR) and feature vector transformation (FVT) methodcalled DRFVT The DR method reduces the dimensionalityof feature vector only the top 119872 low frequency DiscreteFourier Transform (DFT) coefficients are retained These119872coefficients are able to provide all of the abovementionedrequirements and obtain good efficiency in retrieval [13]TheFVT method does the transformation of the original featurevector and generates a new feature vector to solve the problemof noise sensitivity On this basis the similarity betweenthe sketch and the 3D model is calculated [14] To evaluateour method we test our method on the public standarddata set and also compared with other leading 3D modelretrieval approaches [15] The experiments demonstrate thatthe DRFVT method is significantly better than any otherretrieval techniques

2 Our Method

This paper researches on how to get userrsquos retrieval intentionand the feature extraction of 2D freehand sketch Usersexpress their retrieval intention freely by freehand sketch Inorder to realize the 3D model retrieval this paper calculatesthe similarity between 2D sketches feature vector and 3Dmodel projection view feature vector The framework of

the medical freehand sketch 3D model retrieval method isproposed as shown in Figure 1 [16 17]

The 2D freehand sketch framework contains five mod-ules respectively 2D freehand sketch input module 2Doutline representation module DR method module FVTmethod module and 2D sketch feature vector set moduleThe 2D freehand sketch input module would get userrsquosretrieval intention with hand drawing or mouse drawingThe 2D outline representation module extracts the outlineof sketch based on the eight-direction adaptive trackingalgorithm [18] The DR method module reduces the dimen-sionality of feature vector only the top 119872 low frequencyDiscrete Fourier Transform coefficients are retained TheFVTmethod module does the transformation of the originalfeature vector and generates a new feature vector to solve theproblem of noise sensitivity The 2D sketch feature vector setmodule contains the feature vector set generated by the DRmethod and the FVT method

Themedical 3Dmodel framework contains fourmodulesrespectively medical 3D model set module 2D projectionview of 3D model module feature vector extraction moduleand 2D projection view feature vector set module Themedical 3D model set module contains 300 3D model filesof medical organs from the Princeton Shape Benchmark [19]The 2D projection view of 3D model module chooses andgenerates 2D projection views of each 3D model The featurevector extraction module creates feature descriptors whichdescribe the structural information and recognize interiorcontent from 2D projection views of 3D model The 2D

Computational and Mathematical Methods in Medicine 3

3 2 1

4 P 0

765

Figure 2 The point 119875 eight directions

projection view feature vector setmodule contains the featurevector set of 2D projection views of medical 3Dmodel whichwe use to compare with the 2D sketch feature vector set

21 Feature Extraction of 2D Sketch In the general sketchretrieval method the sketches are composed of elementfigures and the relationship between the elementrsquos figures iscomplex For the 3D model retrieval method the sketchesare not expressed by simple element figures because theoutline of the sketch is complicated and has abundant featureinformation it cannot be expressed by simple element figuresTherefore ourmethod firstly extracts the outline of the sketchbased on the eight-direction adaptive tracking algorithmand parameterizes the outline to obtain a complex discrete-time periodic signal whose period 119873 equals the numberof points Secondly we adopt the DR method to reducethe dimensionality of feature vector which obtains goodefficiency in retrieval Finally we propose that the FVTmethoddoes the transformation of the original feature vectorwhich uses the new feature vector to solve the sensitive noiseproblem [20ndash22]

211 2D Outline Representation For the sketches submittedby users this paper extracts the outline of the sketch based onthe eight-direction adaptive tracking algorithm Let 119868(119909 119910)represent the original image [23] and 119862(119909 119910) represent theoutline binary image As shown in Figure 2 we searchthe outline in the 3 times 3 region with point 119875 as centrecounterclockwise

This algorithm selects the first point of 119868(119909 119910) as 119891119894119903119904119905 119901If there are no other points in direction 2 3 4 of 119891119894119903119904119905 119901counterclockwise we save 119891119894119903119904119905 119901 into the outline binaryimage 119862(119909 119910) The algorithm is described as follows

Algorithm 1

Step 1 (initialization) Set the outline binary image 119862(119909 119910)to be null The direction array DI is assigned by DI =

0 1 2 3 4 5 6 7 Let 119889 represent the current direction and119889 = 2

Step 2 Scan the original image 119868(119909 119910) line by line in orderfrom top to bottom and left to right We obtain the startingpoint 119891119894119903119904119905 119901 of the outline and initialize the current point119899119900119908 119901 with 119891119894119903119904119905 119901

Table 1 The normalized DFT coefficients

Invariance Modified coefficientTranslation 1006704119885

0= 0

Scale 1006704119877119898=119877119898

1198771

Rotation 1006704120579119898= 120579119898minus120579minus1+ 1205791

2

Starting point 1006704120579119898= 120579119898+ 119898

120579minus1minus 1205791

2

Step 3 Add 119899119900119908 119901 to the outline binary image 119862(119909 119910)119899119890119909119905 119901 represents the next adjacent point of 119899119900119908 119901 Search119899119890119909119905 119901 according to the direction array order DI[119889]DI[119889 +1] DI[119889+7] We assume the 119899119890119909119905 119901 direction is DI[119894] thenew search direction is 119889 = (DI[119894]+4+1) mod 8Then assign119899119900119908 119901 by 119899119890119909119905 119901 again

Step 4 Judgewhether 119899119900119908 119901 coincideswith the starting point119891119894119903119904119905 119901 if yes then exit otherwise returns Step 3

212 The DR Method Based on Fourier Transform

(1) Method Description The outline of the sketch is closedand cyclical We can use Fourier Transform to describe itscharacteristics First of all we consider the outline of anobject as a discrete-time complex periodical signal 119911 =

⟨1199110 1199111 119911

119873minus1⟩ and define 119911

119897= 119909119897+ 119894119910119897(119894 = radicminus1) where

119909119897and 119910

119897are the real coordinate values of the 119897th (119897 =

0 119873 minus 1) sampled point The signal 119911 is then mapped tothe frequency domain by the DFT [24]

119885119898=

119873minus1

sum

119897=0

119911119897119890minus119894(2120587119897119898119873)

= 119877119898119890119894120579119898

(119898 = minus119873

2 minus1 0 1

119873

2minus 1)

(1)

where119877119898and 120579119898are themodule and the argument of the119898th

DFT coefficient respectively

(2) Invariance Requirement In order to guarantee the trans-lation scaling rotation and starting point invariance wehave to modify the DFT coefficients accordingly When achange is in the position size orientation of the objector the initial point used to parameterize the outline theDFT coefficients are modified [25] In Table 1 we show hownormalized DFT coefficients 1006704119885

119898= 10067041198771198981198901198941006704120579119898 satisfying the

invariance requirements can be obtainedIf one wants to achieve rotation and starting point

invariance the two corresponding modifications have to beintegrated leading to

1006704120579119898= 120579119898minus120579minus1+ 1205791

2+ 119898


2 (2)

For rotation and starting point invariance the derivationis as follows [26ndash28]

We consider the original outline signals 119911 and 1199111015840 where1199111015840 is obtained from 119911 by rotating each point counterclockwise


Table 2 Sample 3D models based on 6 categories

3D model Category 3D model Category 3D model Category

Body Heart Lung

Ovary Kidney Liver

by a constant factor 1205790and by shifting the starting point by 119897

0

positions 1199111015840119897= 119911119897minus1198970

1198901198941205790 The corresponding DFT coefficients

are

1198851015840

119898= 1198851198981198901198941205790119890minus119894(2120587119897

0119898119873)

= 119877119898119890119894(120579119898+1205790minus21205871198970119898119873)

(3)

Thus 1205791015840119898can be expressed as

1205791015840

119898= 120579119898+ 1205790minus21205871198970119898

119873 (4)

In particular when119898 is 1 or minus1 12057910158401and 1205791015840minus1

are

1205791015840

1= 1205791+ 1205790minus21205871198970

119873

1205791015840

minus1= 120579minus1+ 1205790+21205871198970

119873

(5)

By referring to Table 1 it is obtained that

10067041205791015840

119898= 1205791015840

119898minus1205791015840

minus1+ 1205791015840

1

2+ 119898

1205791015840

minus1minus 1205791015840

1

2

= 120579119898+ 1205790minus120579minus1+ 1205791

2minus21205871198970119898

119873minus 1205790+ 119898


2

+21205871198970119898

119873= 120579119898minus120579minus1+ 1205791

2+ 119898


2= 1006704120579119898

(6)

By simply performing the Inverse DFT on normalizedDFT coefficients we obtain a modified normalized signal 1199111015840which satisfies all the requested invariance

1006704119911119897=

1

119872

1198722minus1

sum

119898=minus1198722

1006704119885119898119890119894(2120587119897119898119872)

(119897 = 0 1 119872 minus 1) (7)

(3) Dimensionality Selection In order to concentrate theoutline information of the sketch we use only low frequencyDFT coefficients In particular we keep only the119872 (119872 ≪ 119873)

coefficients whose frequency is closer to 0 The DR methoduses the Bartolini et al [29] spectral characteristics 119864(119872) todetermine the value of 119872 The energy of the signal retainedby the119872 coefficients is defined as

119864 (119872) =

119898=1198722minus1

sum

119898=minus1198722119898 =0

10038161003816100381610038161198851198981003816100381610038161003816

2 (8)

The choice of a suitable value for 119872 has to trade offthe accuracy in representing the original signal with thecompactness and the extraction efficiency

213 The FVT Method The feature vector extraction algo-rithm is as follows

Let the centroid be (1199090 1199100) the distance 119903

119905from the

outline point to the centroid is shown as below

119903119905= radic(119909

119905minus 1199090)2+ (119910119905minus 1199100)2 (9)

Then the distance from each of 119872 outline points to thecentroid (119909

0 1199100) generates the 119872 dimension feature vector

119877 = (1199031 1199032 119903

119872)

The traditional Fourier Transform has certain sensitivityto noise and the noise sensitivity problems will affect thesimilarity accuracy of feature vector Therefore this paperproposes the FVT method which solves the issue of noisesensitivity For the feature vector 119877 = (119903

1 1199032 119903

119872) set

119903119896= min(119903

1 1199032 119903

119872) then generate the new feature vector

1198771015840= (119903119896 119903119896+1

119903119872 1199031 119903

119896minus1)

22 Similarity Comparison From the user input retrievalintention retrieving the userrsquos satisfactory models requirescomparing the feature vector distances between the inputsketch and the 3D model The similarity of models is deter-mined by the differences between feature vectors [30ndash34]

In order to realize the medical freehand sketch 3D modelretrieval this paper compares the similarities of the medical3D model based on Euclidean distance [35] where 119883119884


Figure 3 The medical freehand sketch 3D model retrieval system

Figure 4 The effect of the outline extraction

represent the feature vector of 2D sketch and 2D projectionviews of the 3D model respectively

119863 (119883 119884) = radic

119899

sum

119894=1

(119909119894minus 119910119894)2 (10)

3 Experiments and Results

We implement the medical 3D model retrieval method inC++ under Windows The system consists of a computer

with an Intel Xeon CPU E5520227GHz and 120GB ofRAMOn average each 3Dmodel takes 61 seconds to extractfeatures before the sketch retrieval we can extract and storethe 2D projection view feature vector Moreover the 2Dfreehand sketch takes 006 seconds to extract features Thefreehand sketch retrieval takes 01 seconds in the medical 3Dmodel retrieval system

For performance evaluation we select 300 3Dmodel filesof medical organs from the Princeton Shape Benchmark [18](httpshapecsprincetonedubenchmark) each 3D model


0

02

04

06

08

1

E(M

)E(N

)

100 100010M

Figure 5 The value of 119864(119872)119864(119873) for the heart 3D model

M = 32

Figure 6 The result of the DR method for the heart 3D model

Original human body feature vector

0

05

1

15

2

25

3

Dist

ance

r

604 8 12 16 20 24 28 32 36 40 44 48 52 56 640t

Figure 7 The original human body feature vector


3

Original human body feature vectorWidening human body feature vector

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t

Figure 8 The widening human body feature vector

Original human body feature vectorCutting human body feature vector

3

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t

Figure 9 The cutting human body feature vector

is manually classified based on 6 categories (ldquoBodyrdquo (130images) ldquoHeartrdquo (30) ldquoLungrdquo (25) ldquoOvaryrdquo (13) ldquoKidneyrdquo(22) and ldquoLiverrdquo (18)) The 3D models do not belong to anycategory and were assigned to a default class (62) Table 2shows some 3D models in the data set along with theircategory

In this paper we provide usersrsquo interface with handdrawing or mouse drawing which is convenient for usersto draw and modify the freehand sketches As shown inFigure 3 the left is the tablet which is used for hand drawingor mouse drawing and the right shows 3D model retrievalresults

31 The 2D Sketch Feature Extraction Experiments

311 Extract the Outline of Sketch This paper extracts theoutline of sketch based on the eight-direction adaptivetracking algorithm It is assumed that when the sketch isdrawn the mouse point sequence is stored in 119868(119909 119910) Afterthat 119868(119909 119910) is applied to generate the outline binary image119862(119909 119910) Figure 4 shows a freehand sketch and the outlineextraction

312 The DR Experiments According to the above processwe use the eight-direction adaptive tracking algorithm to


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P

(a) 119875119877 graph on the whole data set

MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P

(b) 119875119877 graph for the query ldquoHeartrdquo

MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P

(c) 119875119877 graph for the query ldquoLungrdquo

Figure 10 119875119877 comparing DR and MAXC method

parameterize 119873 points outline The DR method reducesthe dimensionality of feature vector only the top 119872 lowfrequency DFT coefficients are retained We consider thespectral characteristics 119864(119872) of the data set to determine thevalue of119872

As an example Figure 5 plots the value of 119864(119872)119864(119873)

for the heart 3D model set From the graph when choosing119872 isin [16 64] the retained energy varies from 75 percent to91 percent of the total On the other hand when obtaining119864(119872)119864(119873) = 95 119872 should be equal to 512 The 119872

value will represent 512 outline points and then construct

512 vertices of graph which is not obviously appropriateTherefore choose the smaller 119872 such as 119872 = 32 119872 is apower of 2 we can easily use the Fast Fourier Transform

Experiments of the heart 3D model set show that theeffectiveness of using 119872 = 16 is much lower than 119872 = 32whereas increasing the value of 119872 to 64 does not lead tofurther improvements Thus in this experiment we will use119872 = 32 the result of the DR method is as shown in Figure 6Using 32 generated outline points calculates the distance fromeach point to the centroid (119909

0 1199100) and then generates 32D

feature vector 119877 = (1199031 1199032 119903

32)


DR-FVTDR-NoFVTMAXC-FVT

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P

Figure 11 119875119877 graph for the query ldquoHuman Bodyrdquo

313 The FVT Experiments This paper considers the outlineof the transformed human body such as widening or cuttingthe human body We select the 119872 = 64 value using theDRmethod and calculate the distance from the outline pointto the centroid 119903

119905= radic(119909

119905minus 1199090)2 + (119910

119905minus 1199100)2 The feature

vector 119877 = (1199031 1199032 119903

64) is shown as follows Figure 7 shows

the original human body feature vector Figure 8 shows thewidening human body feature vector and Figure 9 shows thecutting human body feature vector

Figures 8 and 9 show that the feature vectors have certainsensitivity to noise Therefore we transform the existingfeature vector 119877 = (119903

1 1199032 119903

119873) and generate new feature

vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903

119896minus1) to solve noise

sensitive issue

32 Algorithm Performance For evaluation purposes any3D model in the same category of the query is consideredrelevant to that query whereas all other models are con-sidered irrelevant To measure the retrieval effectivenesswe considered classical precision (119875) and recall (119877) metrics[36 37] averaged over the set of processed queries They aredefined as follows

precision =relevant correctly retrieved

all retrieved

recall =relevant correctly retrieved

all relevant

(11)

321 Effect of the DR Method In our first experimentwe compare the DR method with the one based on theextraction from the boundary of the 119872 points having thehighest curvature values hereafter called MAXC [38] We do

not consider methods that reduce dimensionality by simplyresampling the original boundary every 119873119872 points sincethis easily leads to missing significant shape details Resultsin Figure 10(a) clearly show that DR method consistentlyoutperformsMAXC in precision and that it is always positiveeven for higher recall levels A similar trend can be alsoobserved on specific query as shown in Figures 10(b) and10(c)

322 Effect of the FVT Method In order to consider theeffects of the FVT method we compare the DRFVT methodwith the other two methods One method uses the DRmethod and discards the FVT method hereafter called DR-NoFVT The other method uses the MAXC and the FVTmethod called MAXC-FVT For one specific query Figure 11shows the 119875119877 graph for the query ldquoHuman Bodyrdquo andFigure 12 for visualization of retrieval resultsThe experimentresults demonstrate that the FVT method is very useful andtheDRFVTmethod gets better precision than other proposedmethods

4 Discussion and Conclusion

In this paper we propose a novel medical freehand sketch3D model retrieval method DRFVT which uses dimension-ality reduction and feature vector transformation It firstlyprovides a convenient interface to satisfy the userrsquos designprocess and extract the outline of sketch based on eight-direction adaptive tracking algorithm Secondly the DRmethod reduces the dimensionality of feature vector onlythe top119872 low frequency DFT coefficients are retainedThese119872 coefficients are modified so as to achieve the invariance


Figure 12 Results for the query ldquoHuman Bodyrdquo

and then stored in the database Finally The FVT methodis proposed which generates a new feature vector to solvethe problem of noise sensitivity On this basis the similaritybetween the sketch and the 3D model is calculated thus thefinal retrieval results would be presented to the users Theexperiment results demonstrate that our method achievesretrieval results more effectively and efficiently than otherpreviously proposed methods

The future of our work is to develop a user interactionfeedbackmechanismAfter the users submit the query sketchthe system first provides a list of retrieved 3D models Thenthe users can refine the retrieval results by selecting the 3Dmodels which they take as the good results This feedbackmechanism can not only providemore desirable retrieved 3Dmodels to the user but also enhance the user interaction justby making some easy choices It would largely improve theeffectiveness of our retrieval system [39 40]

Competing Interests

The authors declare that there are no competing interestsregarding the publication of this paper

Acknowledgments

This project is funded by the China National Science Councilunder Grant no 61272286 and by Shaanxi Province NaturalScience Foundation of China under Grant no 2014JM8346

References

[1] G Quellec M Lamard G Cazuguel B Cochener and CRoux ldquoWavelet optimization for content-based image retrievalin medical databasesrdquoMedical Image Analysis vol 14 no 2 pp227ndash241 2010

[2] J Kalpathy-Cramer and W Hersh ldquoEffectiveness of global fea-tures for automatic medical image classification and retrievalmdashthe experiences of OHSU at ImageCLEFmedrdquo Pattern Recogni-tion Letters vol 29 no 15 pp 2032ndash2038 2008

[3] B Li Y Lu A Godil et al ldquoA comparison of methods forsketch-based 3D shape retrievalrdquo Computer Vision and ImageUnderstanding vol 119 pp 57ndash80 2014

[4] M Eitz R Richter T Boubekeur K Hildebrand andM AlexaldquoSketch-based shape retrievalrdquo ACM Transactions on Graphicsvol 31 no 4 article 31 2012

[5] J T Pu K Y Lou and K Ramani ldquoA 2D sketch-based userinterface for 3D CADmodel retrievalrdquo Computer-Aided Designand Applications vol 2 no 6 pp 717ndash725 2005

[6] T Funkhouser P Min M Kazhdan et al ldquoA search engine for3D modelsrdquo ACM Transactions on Graphics vol 22 no 1 pp83ndash105 2003

[7] D-Y Chen X-P Tian Y-T Shen and M Ouhyoung ldquoOnvisual similarity based 3D model retrievalrdquo Computer GraphicsForum vol 22 no 3 pp 223ndash232 2003

[8] B Li Y J Lu and H Johan ldquoSketch-based 3D model retrievalby viewpoint entropy-based adaptive view clusteringrdquo in Pro-ceedings of the Eurographics Workshop on 3D Object Retrievalpp 89ndash96 Girona Spain May 2013


[9] F Wang L Lin and M Tang ldquoA new sketch-based 3D modelretrieval approach by using global and local featuresrdquoGraphicalModels vol 76 no 3 pp 128ndash139 2014

[10] D Zhang andG Lu ldquoShape-based image retrieval using genericFourier descriptorrdquo Signal Processing Image Communicationvol 17 no 10 pp 825ndash848 2002

[11] L Yi Z Hongbin and Q Hong ldquoShape topics a compact rep-resentation and new algorithms for 3D partial shape retrievalrdquoin Proceedings of the IEEE Computer Society Conference onComputer Vision and Pattern Recognition (CVPR rsquo06) pp 2025ndash2032 June 2006

[12] J Pu and K Ramani ldquoA 3D model retrieval method using 2Dfreehand sketchesrdquo in Computational SciencemdashICCS 2005 5thInternational Conference Atlanta GA USA May 22ndash25 2005Proceedings Part II vol 3515 of Lecture Notes in ComputerScience pp 343ndash347 Springer Berlin Germany 2005

[13] D Rafiei and A O Mendelzon ldquoEfficient retrieval of similarshapesrdquoThe VLDB Journal vol 11 no 1 pp 17ndash27 2002

[14] L Olsen F F Samavati M C Sousa and J A Jorge ldquoSketch-based modeling a surveyrdquo Computers and Graphics vol 33 no1 pp 85ndash103 2009

[15] T Shao W Xu K Yin J Wang K Zhou and B GuoldquoDiscriminative sketch-based 3D model retrieval via robustshape matchingrdquo Computer Graphics Forum vol 30 no 7 pp2011ndash2020 2011

[16] J Pu and K Ramani ldquoShapeLab a unified framework for 2D amp3D shape retrievalrdquo in Proceedings of the 3rd International Sym-posium on 3D Data Processing Visualization and Transmissionpp 1072ndash1079 Chapel Hill NC USA June 2006

[17] B Li and H Johan ldquoSketch-based 3D model retrieval by incor-porating 2D-3D alignmentrdquoMultimedia Tools and Applicationsvol 65 no 3 pp 363ndash385 2013

[18] M Eitz J Hays and M Alexa ldquoHow do humans sketchobjectsrdquo ACM Transactions on Graphics vol 31 no 4 pp 1ndash102012

[19] S Philip ldquoThe princeton shape benchmarkrdquo in Proceedings ofthe International Conference on Shape Modelling and Applica-tions (SMI rsquo04) pp 167ndash178 2004

[20] Y-J Liu X Luo A Joneja C-X Ma X-L Fu and D SongldquoUser-adaptive sketch-based 3-D CAD model retrievalrdquo IEEETransactions onAutomation Science and Engineering vol 10 no3 pp 783ndash795 2013

[21] S M Yoon M Scherer T Schreck and A Kuijper ldquoSketch-based 3D model retrieval using diffusion tensor fields of sug-gestive contoursrdquo in Proceedings of the 18th ACM InternationalConference onMultimedia (MM rsquo10) pp 193ndash200 Firenze ItalyOctober 2010

[22] C Zou C Wang Y Wen L Zhang and J Liu ldquoViewpoint-aware representation for sketch-based 3D model retrievalrdquoIEEE Signal Processing Letters vol 21 no 8 pp 966ndash970 2014

[23] S Andrews C McIntosh and G Hamarneh ldquoConvex multi-region probabilistic segmentation with shape prior in theisometric log-ratio transformation spacerdquo in Proceedings of theIEEE International Conference on Computer Vision (ICCV rsquo11)pp 2096ndash2103 IEEE Barcelona Spain November 2011

[24] Y Rui A C She andT SHuang ldquoModified Fourier descriptorsfor shape representationmdasha practical approachrdquo in Proceedingsof the 1st International Workshop on Image Databases and MultiMedia Search pp 22ndash23 1996

[25] D Zhang and G Lu ldquoComparative study of fourier descriptorsfor shape representation and retrievalrdquo in Proceedings of the 5th

Asian Conference Computer Vision (ACCV rsquo02) pp 646ndash651Melbourne Australia 2002

[26] I Bartolini P Ciaccia and M Patella ldquoUsing the time warpingdistance for fourier-based shape retrievalrdquo Tech Rep IEIIT-BO-03-02 IEIITBO-CNR 2002

[27] M Kazhdan T Funkhouser and S Rusinkiewicz ldquoRota-tion invariant spherical harmonic representation of 3D shapedescriptorsrdquo in Proceedings of the EurographicsACM SIG-GRAPH Symposium on Geometry Processing (SGP rsquo03) vol 43pp 156ndash164 Berlin Germany 2003

[28] T P Wallace and P A Wintz ldquoAn efficient three-dimensionalaircraft recognition algorithm using normalized fourierdescriptorsrdquo Computer Graphics and Image Processing vol 13no 2 pp 99ndash126 1980

[29] I Bartolini P Ciaccia andMPatella ldquoWARP accurate retrievalof shapes using phase of fourier descriptors and time warpingdistancerdquo IEEE Transactions on Pattern Analysis and MachineIntelligence vol 27 no 1 pp 142ndash147 2005

[30] J M Saavedra B Bustos T Schreck S Yoon and M SchererldquoSketch-based 3D model retrieval using keyshapes for globaland local representationrdquo Proceedings of the 5th EurographicsConference on 3D Object Retrieval (EG 3DOR rsquo12) pp 47ndash502012

[31] M Chaouch and A Verroust-Blondet ldquoA new descriptor for 2Ddepth image indexing and 3D model retrievalrdquo in Proceedingsof the 14th IEEE International Conference on Image Processing(ICIP rsquo07) pp VI373ndashVI376 SanAntonio TexUSA September2007

[32] J-L Shih C-H Lee and J T Wang ldquoA new 3D modelretrieval approach based on the elevation descriptorrdquo PatternRecognition vol 40 no 1 pp 283ndash295 2007

[33] P Papadakis I Pratikakis S Perantonis and T TheoharisldquoEfficient 3D shape matching and retrieval using a concreteradialized spherical projection representationrdquo Pattern Recog-nition vol 40 no 9 pp 2437ndash2452 2007

[34] J Saavedra B Bustos M Scherer and T Schreck ldquoSTELAsketch-based 3D model retrieval using a structure-based localapproachrdquo in Proceedings of 1st ACM International Conferenceon Multimedia Retrieval (ICMR rsquo11) Trento Italy April 2011

[35] C Grigorescu andN Petkov ldquoDistance sets for shape filters andshape recognitionrdquo IEEE Transactions on Image Processing vol12 no 10 pp 1274ndash1286 2003

[36] K Mikolajczyk and C Schmid ldquoA performance evaluation oflocal descriptorsrdquo IEEE Transactions on Pattern Analysis andMachine Intelligence vol 27 no 10 pp 1615ndash1630 2005

[37] J Rocchio and The SMART Retrieval System ldquoExperimentsin automatic document processingrdquo in Relevance Feedback inInformation Retrieval pp 313ndash323 1971

[38] W Y Ma and B S Manjunath ldquoNeTra a toolbox for navigatinglarge image databasesrdquo in Proceedings of the InternationalConference on Image Processing (ICIP rsquo97) vol 1 pp 568ndash571October 1997

[39] PDaras AAxenopoulos andG Litos ldquoInvestigating the effectsof multiple factors towards more accurate 3-D object retrievalrdquoIEEE Transactions on Multimedia vol 14 no 2 pp 374ndash3882012

[40] P Papadakis Content-based 3D model retrieval considering theuserrsquos relevance feedback [PhD thesis]TheUniversity ofAthens2009

Submit your manuscripts athttpwwwhindawicom

Stem CellsInternational

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


MEDIATORSINFLAMMATION

of


Behavioural Neurology

EndocrinologyInternational Journal of



Disease Markers


BioMed Research International

OncologyJournal of



Oxidative Medicine and Cellular Longevity


PPAR Research

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014

Immunology ResearchHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of

ObesityJournal of



Computational and Mathematical Methods in Medicine

OphthalmologyJournal of


Diabetes ResearchJournal of



Research and TreatmentAIDS


Gastroenterology Research and Practice


Parkinsonrsquos Disease

Evidence-Based Complementary and Alternative Medicine

Volume 2014Hindawi Publishing Corporationhttpwwwhindawicom


2D freehand sketch input

Start

Medical3D model set

2D projectionview of 3D model

Feature vectorextraction

Compare similarities 2D projection viewfeature vector set

Results of retrieval

End

2D outline representation

DR method

FVT method

2D sketchfeature vector set

Figure 1 The framework of the medical freehand sketch 3D model retrieval method

For effective retrieval the feature descriptors should be robustto noise and invariant to transformations

To tackle these problems this paper proposes a novelfreehand sketch 3D medical model retrieval method whichcombines the characteristics of dimensionality reduction(DR) and feature vector transformation (FVT) methodcalled DRFVT The DR method reduces the dimensionalityof feature vector only the top 119872 low frequency DiscreteFourier Transform (DFT) coefficients are retained These119872coefficients are able to provide all of the abovementionedrequirements and obtain good efficiency in retrieval [13]TheFVT method does the transformation of the original featurevector and generates a new feature vector to solve the problemof noise sensitivity On this basis the similarity betweenthe sketch and the 3D model is calculated [14] To evaluateour method we test our method on the public standarddata set and also compared with other leading 3D modelretrieval approaches [15] The experiments demonstrate thatthe DRFVT method is significantly better than any otherretrieval techniques

2 Our Method

This paper researches on how to get userrsquos retrieval intentionand the feature extraction of 2D freehand sketch Usersexpress their retrieval intention freely by freehand sketch Inorder to realize the 3D model retrieval this paper calculatesthe similarity between 2D sketches feature vector and 3Dmodel projection view feature vector The framework of

the medical freehand sketch 3D model retrieval method isproposed as shown in Figure 1 [16 17]

The 2D freehand sketch framework contains five mod-ules respectively 2D freehand sketch input module 2Doutline representation module DR method module FVTmethod module and 2D sketch feature vector set moduleThe 2D freehand sketch input module would get userrsquosretrieval intention with hand drawing or mouse drawingThe 2D outline representation module extracts the outlineof sketch based on the eight-direction adaptive trackingalgorithm [18] The DR method module reduces the dimen-sionality of feature vector only the top 119872 low frequencyDiscrete Fourier Transform coefficients are retained TheFVTmethod module does the transformation of the originalfeature vector and generates a new feature vector to solve theproblem of noise sensitivity The 2D sketch feature vector setmodule contains the feature vector set generated by the DRmethod and the FVT method

Themedical 3Dmodel framework contains fourmodulesrespectively medical 3D model set module 2D projectionview of 3D model module feature vector extraction moduleand 2D projection view feature vector set module Themedical 3D model set module contains 300 3D model filesof medical organs from the Princeton Shape Benchmark [19]The 2D projection view of 3D model module chooses andgenerates 2D projection views of each 3D model The featurevector extraction module creates feature descriptors whichdescribe the structural information and recognize interiorcontent from 2D projection views of 3D model The 2D


3 2 1

4 P 0

765






Algorithm 1






0= 0

Scale 1006704119877119898=119877119898

1198771


2

Starting point 1006704120579119898= 120579119898+ 119898


2





⟨1199110 1199111 119911



119909119897and 119910



119885119898=

119873minus1

sum

119897=0

119911119897119890minus119894(2120587119897119898119873)

= 119877119898119890119894120579119898

(119898 = minus119873

2 minus1 0 1

119873

2minus 1)

(1)




119898= 10067041198771198981198901198941006704120579119898 satisfying the



1006704120579119898= 120579119898minus120579minus1+ 1205791

2+ 119898


2 (2)






Body Heart Lung

Ovary Kidney Liver


0

positions 1199111015840119897= 119911119897minus1198970


are

1198851015840

119898= 1198851198981198901198941205790119890minus119894(2120587119897

0119898119873)

= 119877119898119890119894(120579119898+1205790minus21205871198970119898119873)

(3)


1205791015840

119898= 120579119898+ 1205790minus21205871198970119898

119873 (4)


are

1205791015840

1= 1205791+ 1205790minus21205871198970

119873

1205791015840

minus1= 120579minus1+ 1205790+21205871198970

119873

(5)


10067041205791015840

119898= 1205791015840

119898minus1205791015840

minus1+ 1205791015840

1

2+ 119898

1205791015840


1

2

= 120579119898+ 1205790minus120579minus1+ 1205791

2minus21205871198970119898

119873minus 1205790+ 119898


2

+21205871198970119898

119873= 120579119898minus120579minus1+ 1205791

2+ 119898


2= 1006704120579119898

(6)


1006704119911119897=

1

119872

1198722minus1

sum

119898=minus1198722

1006704119885119898119890119894(2120587119897119898119872)

(119897 = 0 1 119872 minus 1) (7)



119864 (119872) =

119898=1198722minus1

sum

119898=minus1198722119898 =0

10038161003816100381610038161198851198981003816100381610038161003816

2 (8)




119905from the


119903119905= radic(119909

119905minus 1199090)2+ (119910119905minus 1199100)2 (9)



119877 = (1199031 1199032 119903

119872)


1 1199032 119903

119872) set

119903119896= min(119903

1 1199032 119903


1198771015840= (119903119896 119903119896+1

119903119872 1199031 119903

119896minus1)







119863 (119883 119884) = radic

119899

sum

119894=1

(119909119894minus 119910119894)2 (10)






0

02

04

06

08

1

E(M

)E(N

)

100 100010M


M = 32



0

05

1

15

2

25

3

Dist

ance

r

604 8 12 16 20 24 28 32 36 40 44 48 52 56 640t



3


0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t



3

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t








MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P










feature vector 119877 = (1199031 1199032 119903

32)



01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















3 2 1

4 P 0

765






Algorithm 1






0= 0

Scale 1006704119877119898=119877119898

1198771


2

Starting point 1006704120579119898= 120579119898+ 119898


2





⟨1199110 1199111 119911



119909119897and 119910



119885119898=

119873minus1

sum

119897=0

119911119897119890minus119894(2120587119897119898119873)

= 119877119898119890119894120579119898

(119898 = minus119873

2 minus1 0 1

119873

2minus 1)

(1)




119898= 10067041198771198981198901198941006704120579119898 satisfying the



1006704120579119898= 120579119898minus120579minus1+ 1205791

2+ 119898


2 (2)






Body Heart Lung

Ovary Kidney Liver


0

positions 1199111015840119897= 119911119897minus1198970


are

1198851015840

119898= 1198851198981198901198941205790119890minus119894(2120587119897

0119898119873)

= 119877119898119890119894(120579119898+1205790minus21205871198970119898119873)

(3)


1205791015840

119898= 120579119898+ 1205790minus21205871198970119898

119873 (4)


are

1205791015840

1= 1205791+ 1205790minus21205871198970

119873

1205791015840

minus1= 120579minus1+ 1205790+21205871198970

119873

(5)


10067041205791015840

119898= 1205791015840

119898minus1205791015840

minus1+ 1205791015840

1

2+ 119898

1205791015840


1

2

= 120579119898+ 1205790minus120579minus1+ 1205791

2minus21205871198970119898

119873minus 1205790+ 119898


2

+21205871198970119898

119873= 120579119898minus120579minus1+ 1205791

2+ 119898


2= 1006704120579119898

(6)


1006704119911119897=

1

119872

1198722minus1

sum

119898=minus1198722

1006704119885119898119890119894(2120587119897119898119872)

(119897 = 0 1 119872 minus 1) (7)



119864 (119872) =

119898=1198722minus1

sum

119898=minus1198722119898 =0

10038161003816100381610038161198851198981003816100381610038161003816

2 (8)




119905from the


119903119905= radic(119909

119905minus 1199090)2+ (119910119905minus 1199100)2 (9)



119877 = (1199031 1199032 119903

119872)


1 1199032 119903

119872) set

119903119896= min(119903

1 1199032 119903


1198771015840= (119903119896 119903119896+1

119903119872 1199031 119903

119896minus1)







119863 (119883 119884) = radic

119899

sum

119894=1

(119909119894minus 119910119894)2 (10)






0

02

04

06

08

1

E(M

)E(N

)

100 100010M


M = 32



0

05

1

15

2

25

3

Dist

ance

r

604 8 12 16 20 24 28 32 36 40 44 48 52 56 640t



3


0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t



3

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t








MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P










feature vector 119877 = (1199031 1199032 119903

32)



01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of



















Body Heart Lung

Ovary Kidney Liver


0

positions 1199111015840119897= 119911119897minus1198970


are

1198851015840

119898= 1198851198981198901198941205790119890minus119894(2120587119897

0119898119873)

= 119877119898119890119894(120579119898+1205790minus21205871198970119898119873)

(3)


1205791015840

119898= 120579119898+ 1205790minus21205871198970119898

119873 (4)


are

1205791015840

1= 1205791+ 1205790minus21205871198970

119873

1205791015840

minus1= 120579minus1+ 1205790+21205871198970

119873

(5)


10067041205791015840

119898= 1205791015840

119898minus1205791015840

minus1+ 1205791015840

1

2+ 119898

1205791015840


1

2

= 120579119898+ 1205790minus120579minus1+ 1205791

2minus21205871198970119898

119873minus 1205790+ 119898


2

+21205871198970119898

119873= 120579119898minus120579minus1+ 1205791

2+ 119898


2= 1006704120579119898

(6)


1006704119911119897=

1

119872

1198722minus1

sum

119898=minus1198722

1006704119885119898119890119894(2120587119897119898119872)

(119897 = 0 1 119872 minus 1) (7)



119864 (119872) =

119898=1198722minus1

sum

119898=minus1198722119898 =0

10038161003816100381610038161198851198981003816100381610038161003816

2 (8)




119905from the


119903119905= radic(119909

119905minus 1199090)2+ (119910119905minus 1199100)2 (9)



119877 = (1199031 1199032 119903

119872)


1 1199032 119903

119872) set

119903119896= min(119903

1 1199032 119903


1198771015840= (119903119896 119903119896+1

119903119872 1199031 119903

119896minus1)







119863 (119883 119884) = radic

119899

sum

119894=1

(119909119894minus 119910119894)2 (10)






0

02

04

06

08

1

E(M

)E(N

)

100 100010M


M = 32



0

05

1

15

2

25

3

Dist

ance

r

604 8 12 16 20 24 28 32 36 40 44 48 52 56 640t



3


0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t



3

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t








MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P










feature vector 119877 = (1199031 1199032 119903

32)



01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




















119863 (119883 119884) = radic

119899

sum

119894=1

(119909119894minus 119910119894)2 (10)






0

02

04

06

08

1

E(M

)E(N

)

100 100010M


M = 32



0

05

1

15

2

25

3

Dist

ance

r

604 8 12 16 20 24 28 32 36 40 44 48 52 56 640t



3


0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t



3

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t








MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P










feature vector 119877 = (1199031 1199032 119903

32)



01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















0

02

04

06

08

1

E(M

)E(N

)

100 100010M


M = 32



0

05

1

15

2

25

3

Dist

ance

r

604 8 12 16 20 24 28 32 36 40 44 48 52 56 640t



3


0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t



3

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t








MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P










feature vector 119877 = (1199031 1199032 119903

32)



01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















3


0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t



3

0

05

1

15

2

25

Dist

ance

r

4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 640t








MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P










feature vector 119877 = (1199031 1199032 119903

32)



01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of

















MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P


MAXCDR

01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

P










feature vector 119877 = (1199031 1199032 119903

32)



01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of


















01 02 03 04 05 06 07 08 09 10R

0

01

02

03

04

05

06

07

08

P



119905= radic(119909

119905minus 1199090)2 + (119910


vector 119877 = (1199031 1199032 119903




1 1199032 119903


vector 1198771015840

= (119903119896 119903119896+1

119903119873 1199031 119903


sensitive issue



all retrieved


all relevant

(11)










Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of




















Competing Interests


Acknowledgments


References
















































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of























































of






Disease Markers



OncologyJournal of





PPAR Research



Journal of

ObesityJournal of
















Research Article A Novel Medical Freehand Sketch 3D Model ...

Documents

Transcript of Research Article A Novel Medical Freehand Sketch 3D Model ...