A study on image segmentation techniques used in color detection

1

ID: 75

A Studv on Tmaee Segmentation Technioues used

in Color Detection

Hasan Alhajhamad

MaGIC-X UTM-IRDA Univcrsiti Tcknologi Malaysia Johor Bahru, Malaysia

[email protected]

Mohd Shahrizal Sunar

MaGIC-X UTM-IRDA Universiti Teknologi Malaysia Johor Bahru, Malaysia

[email protected]

Abstract— to humans, an image is a meaningful arrangement of

regions and objects, whereas to computers, an image is merely a random collection of pixels. This work exploits some of the techniques based on the extraction of the color of an image in the real-time environment. Image segmentation is an intense research area in Computer Vision, however, enhancements or results still on to produce accurate segmentation results for images. Relating with other surveys that compare multiple techniques, this paper takes the advantage to select of the most used techniaue(s). Our study may be helpful for Augmented Reality environment, object detection aud tracking as well as other real-time technologies. Interested reader will obtain knowledge on various categories and types of research challenges in the image-based segmentation within the scope of colored ?nvironn!6nts.

Keywords- Segmentation; Intensity; color; photometry; thresholding; Real-time Image Processing; Augmented Reality

I. Intro duc tio n

D orh'finninrt n f on im ono in tn ronorofo ram A nr ic

1 Ml MVii/lllllk VI Mil IIIIU^V IlliV L’vj.'MI UtV t V^IVI ll_> III WI IUIVIVI VU

to be a part o f content analysis and image under-standing fl]. The process o f Image segmentation is a critical step in many applications. Nowadays, it is fundamentally used in Augmented Reality technology in the cases o f object tracking, registration, and when inserting the synthetic object(s) to the scene. Work on methods and techniques for segmentation had been done thoroughly, but still, results are not yet accurate which makes it difficult to find a method to be called the most

suitable However, the techninne taken on in this naner is to_{' 1 1 I}

scientifically and practically prove in an environment-dependent way to compare between each segmentation method so that the performance can be the judge [2].

Although, the field o f Augmented Reality has come oui io transparence for over one or two decades, work is getting more interesting for researchers; several conferences and seminars have been dedicated to describe the problems o f Augmented Reality and to digest its developments. Registration is basically meant to save the geometrical and the distance o f the annotated objects in the scene. That means, the accuracy o f registration de-pends on object tracking technique which relies on the scene environment. Alas, it is difficult to find a reliable

** .1% * U rt 11« rt«f rt 'I « «rt *Urt

tv/ IVI iv/ w wiiviw II iw iwui uujwvui Ulk UIIUVUICU III II iw

real scene. The first process in most o f the tracking techniques is to define the boundaries, edges, and shapes along the image, which is image segmentation [2].

Image segmentation techniques are divided into two categories; human-guided techniques, also called supervised and unsupervised where segmentation generation is based on

software computing. In other words, unsupervised

s e g m e n ta tio n is w h e r e th e rn m n iita tio n a n a lv s e s a re b a s e d o n_{W x j}

learning algorithm to do the grouping o f common pixels. On the other hand, when the user can sclcct the groups o f common pixels it is then supervised segmentation. That includes the user ability to set the ranges o f how close the similarities in each group io oc [3j.

Many ideas have been presented to develop feasible output for image segmentation. In the next discussion, details o f the different supervised and semi-supervised methods are presented.

II TYPFS OF TMAGF SFGMFNTATION

TECHNIQUES

Defining boundaries, edges, and shapes in an im age

makes it pa<tv to lindprstanH its pnvirnnmpnt that hplns_{J I}

applications in Computer Vision, Artificial Intelligence, CAD and many other areas. For exam ple, locating a m oving object, m otion recognition, detection of suspicious activities, vid eo indexing, human-computer interaction (gesture recognition, eye gaze tracking), vehicle navigation and traffic tracking, also CAD applications like resemble, analyze and m odify parts of objects to do reconstructions or new or im prove m odels, and m odeling o f aesthetic

j . . . ... 1 _.__ ... .1.1 . . ... .1. ut_*M£iu> u:>t: iiuiicriicti (eiUMUii, i i u u u i e ^ l u u ^ i i

surfaces, etc.) [4],

Image segm entation techniques are divided into several categories: Edge-based and Region-based Detection techniques, Partial Differential Equation, Artificial Neural N etwork and Clustering based, and Multiobjective Image Segmentation. In addition to Thresholding Method is also important to be considered. In this paper, study is focused on tw o types, position estim ation m ethods and color estimation methods, comparing each m ethod performance under an environm ent-dependent scope.

2

A. Color Estimation Methods

Classification groups like Grey-level, color, texture, depth, and motion are considered to be essential types for various methods in image based segmentation. Grey-level image based segmentation techniques are the most fundamentally widely used [5], it is generated by controlling thresholds to transform the image data to a binary region map [5]. Simple thresholding technique can be noted as follows:

g (.x,y) = 0 , if/ (x ,y ) < T and

g ( x ,y) = l , i f / ( x , y ) > T (2

Figure 1: two images for two different thresholding regions in the same environment.

For region-based thrcsolding, consider using two

thresholds, T1 < 1 2 , where a region 1 can have a specific range

o f grey levels:

. ..

j ) — V i i V1\

f(x, y) > T 2 a n d g (x ,y ) = 1 if T1 < f(x,y) < T2. (3

Thresholding types and techniques are Simple greyscale

A Z. .. .... .? mmmm - s ?•

LLiiiJbtiUivilli^y nJo«j}uVv Liii'CSiiiJiviisi^, [5],

Attention has been extensive on algorithms for

oon*y>pntoti am Af prt 1 /-\r imorrpo oUliAimU tlia nco /xf* w otr

ievei im age segm entation techniques know ing that they

supplanted many techniques in gray-level image

segmentation. Color Segmentation is crucial for indexing and managing image content. In order to understand color based

image segmentation, it's fundamental to recognize color

representation in terms o f its features; clustering and thresholding. Clustering techniques use the special properties o f colors, color quantization is inseparable as a problem o f

r.lu sterin p p o in ts in t h r e e -d im e n s io n a l s n a r e w h il e in

iiii woiiuiuilig, ji ii li iu i ij iv/ giuV- iCVCi bvgiiiviiluiil/ii ivulivvo iiiU

color level o f an image to color spaces [21].

Although, image texture segmentation techniques are not

still attempts io use iexiuie properties to cluster iexels in an image are growing and showing an interest for researchers [1], [21-23].

SiivTiiislaiiH! hsKui ijii £:“ :in :-:n iasgsiinuiin from in::

u c p u i m a p s , lu ic s p c c i iu the u c lu u iiu ii u l u c p tli iiiap as an

image that provides information on how objects in an image are located, combining color image segmentation technique with the depth information can greatly improve the accuracy o f the segmentation.

Motion based segmentation is the process o f recognizing an cbjcct in a scqucncc o f images that moves dynamically from frame to frame. Motion based segmentation methods aim to partitioning an image into regions based on motion fields o f continuity and the probability o f any parametric motion

model. Few techniques are proposed like Top-down

techniques, Joint estimation, and Grouping o f elementary regions.

As an example on color based segmentation [24], let a color image be denoted as a vector function f*. Color imauc f is usually represented in terms o f RGB format (Red, Green, and Blue), where:

f(.n) = [/R ed (n ) * fGreen (n ) * /B lu e (n )] e R 3 ,n e N

Or it can be noted as indexed palette:

p ( .) e {1, ....,P},C = [c[ ...c j ...Cp]r = [r^ b J eR '’*3

\\J hpro

/R ed(n) : N - > R = {r l,r 2 ...rR} € [0,1]

/G reen(n) : N —> R = {g l,g 2 ,...,g G } e [0,1]

r K iiia tn i • rd U — ( k l hV <= Fn 11

In this example, four stages o f computations are presumed: down sampling, low-pass filtering, color quantization, and

color matching in the L * u * v color space.

Down-sampling is the process to reduce the number o f pixels based on spatial information o f RGB components to

Nl*N2

about ■ M-,-- , where, N ! and N2 are the image size

sublattice noted as SMxM — s u b la ttic e ( N), for every pixel in

every M in both directions (horizontal and vertical):

v (v ) = /( /.) ,_ _ _ = ;

s e S MXM = s u b la ttic e (N ) (5

Then for each o f the three color components o f y ( s ) is to

be low pass filtered by means o f Gaussian FIR so that it adds

smoothness to the image and populates the RGB space with new colors which helps in determining the main colored regions. Next is to apply color quantization method like Heckbert’s minimum variance quantization method [25] for

quantization is used to rcducc the color palette o f an image into a smaller one. The parameter Q represents segmentation resolution for every color space.

Lastly and as a final point is to implement a color matching algorithm in the L*u*v color space; every color space is

identified hy ~:?J55iiring ! •:-'idc~in distances r.oiresnonding to

apprehended color differences.

A coherent and robust real-time video segmentation can be achieved under conditions that will be described in Discussion

111. DISCUSSION

In the previous sections, right the way through the study o f

im age segm entations algorithms focused on the m ulti­ class. single-label setup, where each im age is assigned one of m ultiple classes. Such m ethods arc- used to give one focu sed /sin g le detail of an image. The accuracy of these m ethods are considered to be low comparing to others. M ethods based on variations caused by shadows, shading,

i •: if-« j : ,a~ i •

connected ridges in the chromatic histogram using Kidge based Distribution A nalysis (RAD). It is a real-time unsupervised approach w ith high accuracy results.

---—.. i_ i _ it______

nuiic^ai^ic, uic mijn;i i c i i u i image di^gmwiiuuiun m me ueiu o f Augmented Reality [77]; One important question is: what is the best way to achieve geometry and color for object detection? Or in other words, is there an optimal way for

giM/niL-irv sin-:! citliH gf::CT;u= ii• j: = lli-ii AR :-:ccnc? This

study provides a comparison between the varieties o f image segmentation techniques aiming to select the most appropriate in the scope o f AR scene. Several papers have been devoted to comparative analysis o f either geometry or color segmentation

I— --- J*

For color segmentation, this work depends on Bahadir Ozdemir et. al. [32] parameters to evaluate each method. Next is tn compare the results between them. Rahadir divided the

iii picviSiCui, ic C a iii p c iiu iiiia iiC e , i i i u sicicCiiOii

accuracy.

iTiu \-,Gior Dtiscci ^cftinciitation.

precision =

recalls =

# o f correctly detected objects

if nF rsl! rshiprT': No - FA

= — N 7 r ~ *

# o f c o irectly detected objects # of all obiects in th e im aee

detection accuracy —

Nr v

# o f com pletely detected objects # of all objects in the image

Nc

where FA, MD and MCD are respectively, number of unm atched objects in the algorithm output (False Alarms),

i abie 1 show s the resuits irom each algorithm.

Color estimation m ethods and as it has been discussed above in its section, color based m ethods observations are along these lines: Histogram based patterns m ethods [111, [42 46], [52], these m ethods are already used in AR and have show n good results in terms of accuracy. In [11], im age calibration using four-point m apping and Harris com er detection are proposed to identify different pictures

in an iii-r-K1-. Pi.-itzi-b-ised Hniiliud |24| us-ia palletlz-iid

formats for representing color im ages. Steps taken for the

process are D ow n sampling, Color Quantization

accounting spatial color interactions through low -pass

nliarinu and tl — :; CliiKteriiiii in ih= RGB spgix. This

m eiliod aims io develop unsupervised automatic seiiiag oi the parameters towards color im age segmentation. The method has proved feasibility in terms of accuracy and nrprifiion M ethods based or* interp-ation of rolor and

rexnirf* >l“ KrnpT'!rfi iz i j are p artially su pervised m ulti-class

Multinomial logistic regression w ith Active Learning meth-od [54] is a sem i-supervised segm entation algorithm for high-dim ensional data, class distributions are m odeled

USissg UijvStM.' i H-r activii lei-r

H owever, this method is suitable for high dimensional data. Several Active Contours based segmentation m ethods [6-9], [55-58] have been used in color

Sss5vnf!nt.=tisiTi fsir AR. OHln-er er. =!. [01 m—h-r-H iises rvv::-

a d a p tiv e in itialisatio n of rejpon-based ar+ive m n to n rs A rilvp v h n rm ir based seg m en tatio n m eth o d s hav e p ro v ed

high quality of accuracy although the less percentage of recalls. Scale-Invariance based segm entation [59] is a supervised approach trained classifier used to classify structures o f different classes at all scales The posterior probabilities, outputted by the classifier, is then used to select appropriate scales at all locations. The scope of this algorithm is limited to primitive shapes and abstract image-:- although its high oe curacy results.

Moreover, Texture-based segm entation m etnoas,

Modal Energy of Deformable Surfaces approach [22] based on energy function w hich expresses the local sm oothness

of bp. ini2 2s thst is d^riv^d bv utilizing d6forniHtions

oi =• 3D deioititabltf suiiaue iisudel. The liuuiatio*! of this m ethod is threshold or number of iterations required to im prove the accuracy of the segm entation and being in real-time. Segmentation based on major features in curvcict dom ain approach [60j is a SVfvl :-sCini -stspuryi:-hjd segm entation m ethod used for m edical im ages to classify features like angular second m oment, contrast, correlation, and entropy. Its results arc no in real-time that it takes from 5-12“r-:-"-:;nd.-: in n n s Bislnfi, Amnnir the 0::-U:-r and TVr-ifur^ based segm entation put io use m ethods, Particle Swarm Optimization based method [3] benefited from the use of Multi-Elitist Particle Swarm Optimization (MEPSO) implementation fnr color imacrp Although it uses rnlnr

s n i i i f x i i i r r h ss p ri 5 f» m p r .r s h r .r i. s n ii iVi’iTS” Tf-SilirS i'lT Tni.S

m ethod are un-reliable for low threshold real-time systems.

In addition to Grey-Color, Color, and Texture based segm entation, several m ethods have used Depth fol-Ori] and Motion [1], [f>y-’/3jas a base for segm enting images. Segmentation by follow ing a planar disparity distribution method [68] is the process of partitioning an im age into separate planar regions prior to disparity calculation using a graph cut approach w ithin each segm ent, producing

sm ooth, accurate disparity maps in ordinarily areas. This method aims to find stereo dis-paritv in large and weakly

tcxturcd reg io n s b ased on d ep th iruorm atioii o i the im age.

It is reliable to use this sort of m ethods in the outdoor AR system s know ing the accuracy of this method is high and reliable for real-time. Graph-based Semi-supervised

. ---- :

segm entation aim s to soive the generai im age matching problem using graph theoretic sem i-supervised learning. It is a useful supervised technique for real-time tracking using optical S o w computations. This m eans it is feasible for A S trucking system s accounting its high accuracy of results.

[2] P. Sanchez-Gonzalcz, F. Gaya, A. Cano, and E. Gomez, Str^mcntdtiou and 3D reconstruction approaches for the design flf iar.aroscowic au£=r:':cnicd reality environments,” ■■io-HeJir.a!

pp i.M, n m .

[3] A. van Belle. I. G. Sprinkhuizen-Kuyper, and L. G. Vuurpijl. "Color Image Segmentation by Particlc Swarm Optimization,”

2:H2.

[4] L. Lucchesc, S. K. Mitra, and S. Barbara. “Color Image Segmentation ; A State-of-thc-Art Survey,” Citeseer, vol. 67, no. 2, pp. 207-221,2001.

15] A. Koschan, “Colour Image Segmentation A Survey,” October,

vol. 94, no. October, p. Berlin, Germany, 1994.

IV. i ’UTURL EXPiiCi’ATiONS AND CONCLUSION

o f

aluoiids!!!:-:

e

il

kc

iliuu! finji.ilir. Ntiiu: ::f Use

points o f interests, and so). One challenge is to develop

algorithms for noisy, compressed, unstructured, and

inconstantly illuminated images in order to solve the problem o f stable segmentation. One more challenge, which has been

i:Hn ht- ih " in isfjra iio n o f a ijjo riih n s t' ir’iorm Hiiim

By integrating the image information o f geometry, area color/texture based depth, and motion algorithms into dynamic data. The dynamic data is a map that contains camera’s

poskkiii, dcluib Oil ieaQilci o l euch p ixel, utea ScHtticiiCs

cuiO-description and geometry, and in case oi muiupie images motion that can be achieved by optical flow techniques. Such combination o f image features is crucial for Augmented

R e a l i t y e n v ir o n m e n t f o r c a lib ra tio n an d t r a c k in g o f ca m e r a

a n d tibiccL-i. T o su m u p a[]-U :uciliei, iiic iiiiii is iu d e v is e fa:-,!

image-based area segmentation methods that can detect the targeted areas and compute its pose from a single image taking into consideration o f the dynamic motion in case o f sequence

E«J

[71

[8]

[9]

flO]

[113

fl2]

ilte re v ie w e r io fiiid a n d s e le c t ih e :tio s i su ita b le in e ilio d for h is

implementation. ^3]

This is a part o f a research project by MaGIC-X UTM-IRDA Universiti Teknologi Malaysia. This research is supported by Research Management

Ceiiie; iKTvlL), Uu;vcisili ieiuiulogi Malaysia (UTM)

Science Fund grant (Vot.R.J13000.7282.4S078).

[14]

K. OMigcr. T E-dcter. S. H??ss»«*3s. A. P. Cossdisrschs. ™d A Mcnms. A novel approach oj initialising region-based acme contours in noisy images by means o f higher order statistics and dissimilarity. 2010. pp. 477-482.

M. Jung. G. Peyre. and I.. D. Cohen. “Nonlocal active contour?.,” ■Vi/iiv .hturtllli I in ilnitvmv .W ru m vni *i mi i ini i i i / 7 - i i i ') 2012.

D. A. G. Baswaraj and D. P Premchand, “Active Contours and ■ /■-: * v : T 1 C :::: ;T;:::' A:: ” Cr’::::::: Journal or Csmsmer Science and i'echnoiogy. voi. i2. no. 11 - i.

2012.

K. Ohliger. T. Edeler. S. Hussmann, and A. Mertins, A novel position and shape aaap-i\:e inidalisadon ofreglon-hased acrhe

7F.FF ?!!! ! p:: h i :■*.!

A. R. C. Paiva and T. Tasdi/en, “Fast semi-supervised image segmentation by novelty selection," in Acoustics Speech and

Conference on, 2010. pp. 1054-1057.

L. Shafarcnko, H. Pctrou, and J. Kittlcr, "Histogram-based segmentation in a perceptually uniform color space.,” IEEE n-ansacnons on Image processing: apublicadon of the lliiiL Signal Frvizz&i'tg Svviviy, vu:. 7, ;:u. v, pp. ; 35-1 S, jl;;;. ; 998.

N. Brewer, N. Liu, and L. Wang, “Guided informative image partitioning," Structural, Syntactic, and Statistical Pattern

7;'-.\?: 7. SOU*I

L. Sankar and D. C. Chandrasekar, “SEMI SUPERVISED IMAGE SEGMENTATION USING OPTIMAL HIERARCHICAL CLUSTERING BY SELECTING

iNTLRiiS i iiU lUiGlUN AS 1:1UUR IWr OiiMA i iON /' Journal o f Global Research in Composer Science, vol. 2, no. • •, pp. 1 -5, 2011.

J. Mitcran, S. Bouillant, and E. Bourennanc, "SVM

■ jf- . • !<ySifli::;- .::: improvcU ii>pvrrcciaiigiw-basvd ntvibvU.” reai-iime imaging, voi. 9, no. 3, pp.’ 179-188,2003.

[1] D. Crcmcrs, M. Rousson, and R. Deriche, “A Review of Statistical Approaches to I^evel Set Segmentation: Integrating Color, Texture, Motion and Shape,” International Journal o f

[15]

[16]

N. Houhou, X. Bresson, A. Szlam, T. Chan, and J.-P. Thiran,

"Stenu-supcrvised based t s non-k-rai vcniiiHHHs

iiiiii-Cut,” oCuic CtjjuCir- unu r ui luittinui :vic irttiuS in v.Otrtpiiici Vision, pp. 112-123,2009.

C. Z. C. Zhang, and X.-S. H. X.-S. f SfiTi' rxifi:t::ifi:r;y. ::::: :::: :: ::r: segmentation. 2wS.

[17]

[18]

[19]

[20]

[21]

[22]

[23]

m

[25]

[26]

[27j

[28]

[29]

Z. Zhcnglao, Y. Xiongyi, H. Liuqian, and W. Dc, Fast capsule image segmentation based on linear region growing, vol. 2.

2011. fH>. vv-10.1.

H. Durrant-whyte and T. Bailey, “Simultaneous Localisation and Mapping ( SLAM ): Part I The Essential Algorithms,” History,

vol. 13, no. 2, pp. 1-9,2006.

T. Dailey and IT. Durrant-tvhytc, “Sintultaneous Localisation and Mapping ( SLAM ): Part II Slate of the Art,” Computational Complexity, vol. 13, no. 3, pp. 1-10, 2006.

n. say. •. •

features,” Grnqma- Vision -EC

D. E. Ilea and P. F. Whelan, “Image segmentation based on the integration of colour-tcxturc descriptors—A review,” Pattern Hlujnyiiiili.-i. Visl 4-1. 10- i I. j:!": 1 4 i

M. Krinidis and I. Pitas, “Color texture segmentation based on the modal energy of deformable surfaces.,” IEEE Transactions on linage Processing, vol. 18, no. 7, pp. 1613-1622, 2009.

M. Mirmehdi ami M. Petrou, Segmentation o f color textures, vol. 22, no. 2. 2000, pp. 142-159.

!.. Titcch-asc stk? P. V. M itra, “ An a ’^nrithrn fo r f - j.i segmentation •-■i uuagcs.' i t t h . v a . \y. ’y'y'6

P. Hcckbcrt, Color image quantization forfram e buffer display,

vol. 16, no. 3. ACM, 1982.

A. iicahdad: H3il W. Sc-isdcsc. riVH-inspired sssrsssh ja r

;:iuluuin:::. 2007.

W.-X. K. W.-X. Kang, Q.-Q. Y. Q.-Q. Yang, and R.-P. L. R.-P. Liang, The Comparative Research on Image Segmentation

135J

[-■-j

[37]

[38] [34]

[40]

[41]

[42]

z,. Wang, K. iioesch, and C. uinzier, Quantitative Comparison o f

Segmentation Results from ADS40 Images in Swiss NFI. IEEE. [43]

2011, pp. 147-151.

[44]

L Yang, F Albregtsen, Tor L Onnestead, and I’ Gr Ottum, “A

supervised approach to the evaluation of image segmentation [46] methods,” in Computer Analysis o f Images and Patterns, V

S. Osher and N. Paragios, Geometric Level Set Method'; in Imaging, Vision, and Graphics. Springer. 2003, pp. 79-99.

N. l’aragios and K. Denche, "Coupled geodesic active regions tor image segmentation: A level set approach,” Computer Vision—

ECCV 2000, pp. 224-240.2000.

5.-Y. W. S.-Y. Was and W. h. iii-i-iiL:. i t■■■■msnii: rsaisn £i*ivhis, vol. 12, no. 9. 2003, pp. T007-l6?5.

Z. Lin and B. Lu, The pupil location based on the Laplace operator and region-growing. IEEE, 2011, pp. 5170-5172.

Y. Chang. “Fast image region growing,” Image and Vision Computing, vol. 13, no. 7, pp. 559-571, 1995.

C H C. Hiisn" O-1 f> T.j«. and X I. Y. C fikr imarr

sastmssiaiion by sccaal ration arvmng anil rcmon mcminie,

vul.

2. IEEE, 2010, pp. 533-536.

J. Beaulieu and R. Touzi, Mean-shift and hierarchical clustering fo r iccinredpc-l-n—etric SAP.;•-■-=c sc^cnss-ioz/c'assiftcction.

20? 0. pp. 251-5-2522.

K. Ohkura, H. Nishizawa, T. Obi, A. Hasegawa, M. Yamaguchi, and N. Ohyama, “Unsupervised image segmentation using hierarchies! ■h-ste-rln-?.” Optlm! ftt-vS-w, vft*. 7. no. 3. pp. '93-

■ :«. ii.w.

A. Harimi and A. Alimadyfard. Image Segmentation Using Coirelative Histogram Modeled hy Gaussian Mixture. Icce, 2009,

T. K. Ganga and V. Karlhikcyani, Medical image segmentation using multi resolution liiitogram, vol. 6. IEEE, 2011, pp. 267- 269.

G. Thomas, 2008.

[45] D. Wcilcr and J. Eggcrt, “Multi-Dimensional Histogram-Based

.-7—

K. Bhoyar and O. Kakde, “COLOR IMAGE SEGMENTATION BASED ON JND COLOR HISTOGRAM,” Image Processing,

[30] P. R. Marpu, M. Neubert, II. Herold, and I. Niemeyer. "Enhanced [47] evaluation of image segmentation results,” Journal o f Spatial

Science, vol. 55, no. 1, pp. 55-68,2010.

Z. L. Z. Liu, X. F. X. Fan, and F. L. F. Lv, Segmentation o f infrared image using support vector machine, vol. 2. IEEE, 2010, pp. 455-458.

115 Ij M. I'olas. li. Zhang, -nil M. ih. An evaluation nt-xtnc ibr segiiisstawoa of iiiukiple objects,” :'~n

vol. 27, no. 8. pp. 1223-1227,2009.

[32] B. Ozdemir, S. Aksoy, S. Eckert, M. Pesaresi, and D. Ehrlich, Tor ionttancc measures tor o-btcci detection cvaiua-ion." Pa-teca& lthi* Liiii:*, vol. 31, so. SO, pp. ! I2S-S 137, 2010.

[33] R. Jaulmes, E. Moline, and J. Obriet-Leclef, “Towards a quantitative evaluation of simultaneous localization and mapping methods." m Csnsrc-i Architecture oi Jiobots xc-ionai caafsrs-xs.

200S.

[4oj W. H. r . W. H. fcna. L. Z. L. Zhuana. iL H. K. Hona, and Z. i:.

Z. ?e::g, 7cu;‘m**c : m&.'.'xxxxim.'i ulgvriSJim viutvsi o:: Nonsubsampled Contourlel Transform and SVM. 2010, pp. 2712- 2716.

[4vj X.-Y. Wang, X.-j. Zhang, ii.-Y. Yang, and j. lia, A pixel-based

COiOf i:VUigc

Uoi;:£ 5UppOit VcCiOf

ui;d

fuzzy C-means..” Neural Networks, vol. 33C. pp. 148-159.2012.

150] S. Li and Y. Li, An application o f linear SVM to fingerprint Ustsgc siiritssusncn. UiL'L, 2011, pp. :.-~=4-vv7.

[51] L. L. L. Liu and T.-yong W. T.-yong Wang, Algorithm ofTi

Segmentation Combining FCM and FSVM. 2009.

D. Bcicr, R. Billert, B. Bruderlin, D. Stichling, and B. (69] Kleinjohann, Marker-less vision based tracking fo r mobile

sufansssed re sa i? . TEEn C o isp s t. S i x , 2f=XV nrt. 2-.1.

F.. Vazquez, J. van de Weijer, and R. Baldrich, “Image [70] segmentation in the presence o f shadows and highlights,”

Computer Vision—ECCV 200/i, pp. 1-14, 2008.

L71J j. L. J. Li, J. M. OioucK-DiiS, and A. P'asa, Suni-supxnisid

hyperspectral image segmentation, vol. 4. TEEE. 2011. pp. 1855- 1858.

M . O. if :.Ti.T. iU'SviriiSg xcciuzic-n in aagm&iii'd rzxiiiy: a COiituiii bas-cd uppiOuJi nuhoiii 3D i ut-O-au uiliO;:. IEEE Comput. Soc, 1997, pp. 9 1 -96.

[73] J. Platonov and M. Langcr, Automatic contour model creation out oi sosvsonal CAD modzi- k-r markerhos Attsmented sicaiiiv. voi. 5. no. 3. U ii, 2007, pp. ! -1.

174] A. Cassinelii, A. Zerroug, Y. VVatanabe, M. Ishikawa, and J.

Angcslcva, "Camcra-lcss Smart Laser Projector,” in ACM

SlCU & ii-si2010Emerging TschtsoU/giss,2010. pp. :::1 . . . .

I • “ I

X. Zhang, M. Gunther, and A. Bongers, “Real-Time Organ Tracking in Ultrasound Imaging Using Active Contours and Conditional Density Propagation,” in Medical Imaging and Juans!ties Hsaiiiv. 20iv. voi. t-326. cc-. 2io-204.

■

Y. Li, D. M. J. Tax, and M. Loog, “Supervised Image Segmentation with Scale Invariance,” image, vol. 7, p. 8.

F?71 i . Vun. "oemi supcjviovu UiiiusounU iuiuxc ocxiuciiiuiioii iJmeU on Direction Energy and Texture Intensity," Appl. Math. vol. 6, no. 3, pp. 737-743,2012.

C. C ids and A. A Alstss.Depth Assi& edQ bkciS~~.~i-.ti~-Aiuiri-View Video. iccc, 2iXfH;'pp- J 83-1 SB.

M. M. Chang, A. M. Tckalp, and M. I. Sczan, An algorithm fo r simultaneous motion estimation and scene segmentation, vol. 6, no. ■;■. teee, iv?7. r=n- i32«-i.r*-.i.

A. J. Patti, Motion-based video segmentation with boundaiy refinement. leee, 2010, pp. 1138-1141.

A. iL Muniour- und A. M-u-.ii-. -ip&iaisieini spsce-dme isaiian

■izgiiiiiiiasiosi o f UiS&gd SajuiriixS by SSIpii:tuii, VC-’:. 2. I««, 2002, p. 11-265-11-268.

P. D. Smet and D. D. Vleeschauwer, Motion-based segmentation using if ihresnxided naTgins sirsiegy ixiier-ked segmeni-, voi. 2. IEEE CO::':pl;S. SiC, :???, pp. 2—5.

I. Patras, E. A. Hendriks, and R. L. Lagendijk, An iterative motion eslimation-segmenlalion method using watershed segments, vol. 1;. iim . pp. U-4.

N. Huang, Graph based semi-supen-ised learning in computer vision. ProQuest, 2009.

A. J. Davison, I. D. Reid, N. D. MMtnn, 2nd O. Stassc, “MonoSLAM: real-time single camera SLAM.,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 29, no. 6, pp. 1052-1067, 2007.

P.. N-«ck and P. Nlclscn. "Scnsi-supcrviscd ststisiicai r££-lr:n refinement for color image segmentation," Pattern Recognition,

vol. 38, no. 6, pp. 835-846,2005.

Kn!5van!i. H!>shan?, and Mohd ShahriTal Sunar. "Realistic Rea!- i line OuUiooi RvSKiviiiiKill AuicuKsiisU Kvuiiiv. ' riuo one y. :iy. 9 (2014): el08334.

[72]

E. Mirante, M. Georgiev, and A. Gotchev, A fast image segmentation algorithm using color and depth map, vol. 0, no. 1.

M. Pardas, Video object segmentation introducing depth and motion information, vol. 2. IEEE, 1998, pp. 637-641.

j. risassdsz asd i. /-rind-:, image segna-niaik-n combining

fiigKlfi dvpili uiid Ol-jCCifia&uuS, Viii. 5.1EEE CiiVipUt. SiC, 2000, pp. 618-621.

C. S. W. C. S. Won, K. P. K. Pyun, and R. M. Gray, Automatic

obiec; in ims~es ~i;h ic-~- decth c-i He id. voL 3. no.

iii. TEEE, 2002, pp. S05-S0S.

A. Yoonessi and C. L. Baker, “Contribution of motion parallax to segmentation and depth perception.,” Journal o f Vision, vol. 11,

E. Francois and B. Chupcau, Depth-bused segmentation, vol. 7. no. 1. 1997, pp. 237-240.

Ap p e n d i x III

Table 1: a summarytableforcoiorestimationbasedcmimagbsegmentationmethods.

N o . M o t l i o d \ T e c h n i ;; w : \ A l g o r i t h i n O i i t K g o r y P a r ; : . r r ; : t e r l R ; i i : ! - . r n e l e r 2 F' : u a m e t e r 3 i 'o m m e n l

C o ! a r E s t i m a t i o n

M e t h o d s

P r e i : i ;;i o n R e t a i l A c c u r a c y

1

1' i d i g r a m b a s e d p a t t e r n s [ 1 1 ] , | ■1 2 —4 6 ] ,

...

C o l o r b a s e d 1 1 u s e d a l r e a d y i j i t h e A R e n v r o n m e n t

2 1 ' : o : l - b a s e d [ 2 4 ] C : i l o r b a s e d 1 1 9 2 %

a i m s t o d e v e l o r t a u t o m a r i t : s e t t h i u :

o f t h e p a r c . n : :: : e r s t o w a r d : ; j n s u p e r v i s e d c o l o r i m a g e

s e g m e n t a t i o n .

3

b a i l e d o n i n t e g r a t i o n o f c o k u t e x t u r e

c.i: -:i; r i p t o r s [ 2 1 ’

C o l o r b a s e d 0 . 5 O -ix

j

i d g i v e o n e M o u s e d / s i n g l e :i.:: r a i l o f a n i r t . a i t e

4

b u o n v a r i a l i o n 5 c a u s e d b y s h a d o w s ,

s i : i : i i i ; l i n g s , a n d h ;.;l i l i g h t s [ S 3 ]

C o l o r b a s e d 1 1 0 | 't % r e a l - t i m e . u t i s v o e r v i s e d , u s i n g h i s t o g r a m i : t f o r m a t i o n

5

M o d a l E n e r g y o f D e f o r m a b l :: S u r f a c e s

' f e ’u r e b a s e d 0 . 6 t: o . e i s ( 6 %

r e q u i r e s t o s r t j i e i f y t h e m n i t s r o f i t e r a l i o t t s t o i m p r o v s

t h e a c c u r a c y i:-l: s e g m e n t a t i o n

6

U s i n g m u l t i n o i x a l l o g i s t i c r > ! : ; j ; r e s s i o n

v.v 11: A c t i v e L e a :i i n g [ 5 4 ]

G i : : y - l e v e l a n d C o l o r

l:iiii!i’! ! d

0 . 6 C: p A W / o s u i t a b l e f o r h i ji b d i m e n s i o n a l d a t a

7 A c t i v e C o n t o u r * ' : i . « s c d [ 6 - 9 ] , ' 5 S — 5 8 ]

G r ; : y - l e v e l a n d t” o l o r

l » i x l

O O .C i:- 8 .5 %

f o r b o t h p < i s i ! o n a n d s h a r K : a d a p t i v e i n i t i a l i z a t i o n o f

r e g i o n - b a s e d a c t i v e c o n t o u r s

8 S o u l s - I n v a r i a n t :: b a s e d [ 5 9 ]

G t ; : y - l e v e l a n d

T : : : - : r u r e b a s e d

1 1 M t %

' C h e s c o p e o f I I I i s a l g o r i t h m i:- l i m i t e d t o p r i m i t i v e s h a p r e

i m a g e s

9

t ' i ! - : i : d o n m a j o r f e a t u r e s i n i : u r v e l e t

c i o n : a i n [ 6 0 ]

G r j i y - l e v e l a n d

T u x u r e b a s e d

0 . 5 0 . 4 7 y / o

u s e d f o r m e t l i e a l p u r p o s i : , r e s u l t s a r e n : : i n r e a l - t i m i :

t t a k e f r o m 3 - 1 ^ s e c o n d s )

1 0 I ' ^ i t i c l e S w a r r r . o p t i m i z a t i o n b u s e d [ 3 ]

C o l o r a n d [ t e x t u r e

b i : i : i o d

1 1 9 :5 %

M ' o r s t r e s u l t ! ! ;t::>s u n r e l i a b l e I t : r l o w t h r e s l i o l d i n r e a l - t i m i :

s y s t e m s

1 1

b y f o l l o w i n g ;i p l a n a r >;l s p a r i t y

c . i i : i : r i b u t i o n [ 6 J!|

" 3 n t h a n d ' [ t e x t u r e

b i : i i i 'i : d

1 1 s o %

i : i m i s t o f i n d s t e r e o d i s p a r i t y i n l a r g e a n d v v e a k b .

t : : : x t u r e d r e g i o n 5 b a s e d o n t it : |: t h i n f o r m a : i o n o f t h e i m a g e

1 2

g ;r a .| :t i b a s e d s o n : i - s u p c r v i s e d l e a r n i n g

[■■•»]

! ' . - l l i o n a n d D e p t h

b i w s d

1 1 9 0 %

s u p e r v i s e d i:e ::l t n i q u e f o r r e e l - t i m e t r a c k i n .; u s i n g o p t i c a l

f l o w c o m p i t a c o n s