A Novel Fast Otsu Digital Image Segmentation Method

Transcription

1 The Inenaonal Aab Jounal of Infomaon Technology, Vol. 3, No. 4, July A Novel Fas Osu Dgal Image Segmenaon Mehod Duaa AlSaeed,, Ahmed oudane,, and Al El-Zaa 3 Depamen of Compue Scence and Dgal Technologes, Nohumba Unvesy a Newcasle, UK College of Compue and Infomaon Scences, Kng Saud Unvesy, Saud Aaba 3 Depamen of Compue Scence, eu Aab Unvesy, ebanon Absac: Dgal mage segmenaon based on Osu s mehod s one of he mos wdely used echnque fo heshold selecon. Wh Osu s mehod, an opmum heshold s found by maxmzng he beween-class vaance and he algohm assumes ha he mage conans wo classes of pxels o b-modal hsogam (e.g., foegound and backgound). I hen calculaes he opmal heshold value sepaang hese wo classes so ha, he beween class vaance s maxmal. The opmum heshold value s found by an exhausve seach among he full ange of gay levels (e.g., 56 levels of nensy). The objecve of hs pape s o develop a fas algohm fo he Osu mehod ha educes he numbe of seach eaons. A new seach echnque s developed and compaed wh he ognal Osu mehod. Expemens on seveal mages show ha he poposed Osu-Checkpons fas mehod gve he same esmaed heshold value wh less numbe of eaons hus esulng n a much less compuaonal complexy. Keywods: Image hesholdng, osu mehod, opmzed seach echnque. Receved Apl 3, 03; acceped June 3, 04; publshed onlne Augus, 05. Inoducon Dgal mage segmenaon s a fundamenal sep n many mage analyss applcaons and s a vey ccal ask. One can defne mage segmenaon as a paonng o cluseng echnque used fo mage analyss. In anohe wods, s a pocess of subdvdng an mage no s consuen egons o objecs as pa of he analyss pocess [4, 7, 5]. Image segmenaon algohms ae geneally based on one of wo basc popees of nensy values: Dsconnuy and smlay. In he fs ype, an mage s paoned based on sudden changes n nensy (e.g., gay level), whle n he ohe ype, an mage s paoned no egons ha ae consdeed o be smla based on cean cea [7, 5]. Osu mehod s one of he wellknown hesholdng mehods and nvolves an eang pocess hough all he possble heshold values and calculang a measue of spead fo he pxel levels each sde of he heshold,.e., he pxels ha fall n ehe foegound o backgound. The am s o deemne he heshold value whee he sum of he foegound and backgound speads s a s mnmum snce he heshold value havng a maxmum beweenclass vaance has also a mnmum whn-class vaance. Theefoe, s a good alenave o use he beween-class vaance fo fndng opmum heshold snce eques less compuaonal complexy [5, 6, 8]. In addon, hee does no exs a sngle mage segmenaon algohm, whch can gve he bes esul fo evey dgal mage. Dependng on o he ype of he gven mage applcaon he mos appopae appoach s o be chosen o acheve he bes segmenaon esul. In some applcaons, he pocessng me (o compuaonal complexy) s an essenal ssue. Fo example, n bomec auhencaon and vefcaon sysems hs s a ypcal consan n addon o an effcen segmenaon. The mehod poposed n hs pape s amed o develop a fase mehod fo esmang he opmal heshold value n Osu mehod by mnmzng he numbe of seach eaons along possble heshold values. Ths s acheved by selecng a sub-ange of gay levels aound an nal esmaed heshold value whle sll keepng he keepng he qualy of he segmenaon pocess nac. The pape s oganzed as follows: Secon noduces he segmenaon pocess n geneal and hesholdng echnque n pacula. Secon 3 pesens a bef noducon on Osu mehod of hesholdng. The poposed echnque s explaned n secon 4 whle secon 5 s concened wh he expemenal esuls caed ou. Secon 6 concludes he pape.. Image Thesholdng Dgal mage segmenaon has been a majo eseach opc n dgal mage pocessng esulng a vaey of mehods and algohms developed and mpoved fo mage segmenaon angng fom geneal segmenaon mehods o aloed fo cean mage ype o a cean applcaon [, 7]. Each mehod has a dffeen appoach n defnng wha chaacezes a good segmenaon and uses a dffeen echnque o fnd he opmal segmenaon [3, 4]. Image

2 48 The Inenaonal Aab Jounal of Infomaon Technology, Vol. 3, No. 4, July 06 hesholdng s one of he man appoaches of segmenaon and Osu hesholdng mehod s one well-known hesholdng echnque. Image hesholdng s consdeed he eases way o segmen an mage. Alhough, seems smple he poblem of choosng a good and an accuae heshold value s a dffcul ask. Thesholdng s mos commonly used fo sepaang objecs fom he backgound [3, 9, 0,, 5]. The mos wdely used hesholdng echnque uses he gay level hsogam. When an mage, f(x, y) s composed of dak objecs on a lgh backgound, hen he foegound s clealy dsngushable fom he backgound. In hs case, he mage hsogam wll be bmodal so ha, he heshold value wll le n he valley of he hsogam ensung ha he objecs can be exaced by compang pxel values wh a heshold T. If any pxel (x, y) fo whch f(x, y) T s consdeed as belongng o he objec class, ohewse, belongs o he backgound class [7, 8]. Unfounaely, hs s no he case n mos mages. In addon o hsogam based hesholdng echnques, seveal echnques have been poposed, yng o fnd a way of choosng he bes value of heshold T ha wll esul n an accuae segmenaon. Osu mehod s one of he wdely used hesholdng mehods; Osu s mehod nvolves eang hough all possble heshold values and calculang he weghed sum of whn-class vaances of he foegound and backgound pxels. The am s o fnd he heshold value whee he sum of whn-class vaances s a s mnmum. 3. The Osu s Mehod e he pxels of a gven pcue be epesened n gay levels [,,..., ]. The numbe of pxels a level s denoed by n and he oal numbe of pxels by N=n +n +...+n. The pobably dsbuon [8,, 6]: n p =, p 0, = p = N Now, when we classfy he pxels no wo classes C 0 and C (backgound and objecs) by a heshold a level, C 0 denoes pxels wh gay levels [,,..., ] and C denoes pxels wh levels [+,..., ]. Then, he pobably of class occuence and class mean especvely ae [, ]: And µ ω = p ( C ) = p = ω( ) 0 0 = ω= p ( C ) = p = ω( ) = + p µ ( ) µ = = = 0 ω ω( ) 0 p ( \ C ) = = 0 p = p ( \ ) C = = = + = + ω µ T µ () ω() () () (3) (4) (5) Whee ω and µ ae he 0 h and s ode cumulave momens of he hsogam up o h level and ae defned as follow: And ω( ) = p = µ ( ) = p And µ s he oal mean level of he ognal mage whch s compued as: = µ = µ ( ) = p T = Fo any ; he followng elaon s vald: µ ω + µω = µ, ω + ω= 0 0 T 0 The class vaances and oal vaance ae gven by: = ( 0 0) p ( \ C ) = ( 0 0 ) = = µ µ p ω = ( ) p ( \ C ) = ( ) = + = + µ µ In ode o evaluae he goodness of he heshold (a level ), whn-class vaance and beween-class vaance ae used as measues of class sepaably. They ae ae defned, especvely as follows: w = ω ω = 0( 0 T ) + ( T ) = 0 ( 0) ω µ µ ω µ µ ωω µ µ And he followng elaons hold: η= T T = + W The poblem s hen educed o maxmze one of hese ceon measues. I s noced ha s dependen on s ode sascs whle depends on he nd ode sascs. Theefoe, s he smples measue wh espec o. Thus, η() s adoped as he ceon measue o selec he bes heshold value () whch s deemned by a sequenal seach by usng he followng funcons []: W () η() = [ µ Tω () µ ()] () = ω T ( )[ ω() ] Snce, T s no a funcon of heshold (), hen he opmal heshold should be he value whch maxmzes ( ). Thus he opmal value heshold (* ) s compued wh he followng equaon []: * ( ) = max ( ), < 0 p ω (6) (7) (8) (9) (0) () () (3) (4) (5) (6) (7) (8)

3 A Novel Fas Osu Dgal Image Segmenaon Mehod Poposed New Fas-Osu Mehods (Osu- Checkpons) In Osu mehod he value of ( ) s compued fo each, 0. Once hs done, he mnmal value of ( ), say ( * ) s seleced fom hesee values and * s hen consdeed as he opmal heshold value []. y usng Equaon 8, fo each he compuaonal effo fo calculang ( ) s boundedd by c, whee c s consan. y consdeng he me complexy noaon [6], le f() be he me complexy fo o calculae ( ), heefoe, f() c,.e.., f()=o(). Fo 0, eques O( ) cycles o calculae all ( ) s [5]. Ou am s o speed up he pocess of Osu hesholdng by developng a fas eave Osu hesholdng algohm. We speed up he pocess by eavely mnmzng he ange of possble heshold values unl he ange eaches s mnmum lengh, whch s 3 gay level. Fo a bmodal hsogam, n mos of he cases he opmal heshold esmaed by Osu mehod should no be fa fom mage global mean mu T. Fgues and show examples of an mage and s hsogam. Consdeng hs, we can naow he seach ange o a cean dsance aound he local mean. Howeve, snce hs s no he case n all mages, nsead of buldng he seach ange only aound he global mean; we popose o se ou nal seach ange o he full gay level ange. In ha ange we hen se hee checkpons, choose he one wh he maxmum eween Class Vaance (CV), hen se as he nal esmaed heshold T n. Aound hs esmaed heshold value, we wll buld he new seach ange and epea he same pocess of seng checkpons and naowng he seach ange unl we each one of ou soppng condons:. The CV a he seleced nal heshold T n has CV lage han he wo gay levels aound (T n -, T n +), n hs case we wll se opmal heshold o T n and ex ou seach pocess.. The ange become only hee gay levels lengh, a hs pon we wll se he opmal heshold o he one of he 3 gay levels ha has he maxmum beween class vaance CV and ex ou seach pocess. Fgue 3 depcs an example of an mage and s hsogam showng he phases of ou seach pocess. a) Ognal. b) Osu-T=9. c) Hsogam. Fgue. Ognal, hsogam and segmened mage of skn cance 5. a) Ognal. b) Osu-T=90. Fgue. Ognal, hsogam and segmened mage of ensen. a) Ognal. c) Phase. Fgue 3. Ognal mage and sages fo fndng opmal heshold fo mage of ensen. Now, n ou nal sage we se he nal check pons o he global mean of he full hsogam as he md-checkpon cp and wo local means of he wo pas of he hsogam afe splng no wo pas on each sde of hs md-checkpon. Ths esuls n 3 nal checkpons (cp, compue ( (cp ), (cp), (cp 3 )), especvely. The hee nal checkpons ae defned, especvely as: cp = µ = jp T 3 cp = cp j= j = cp+ These check pons dvde he mage hsogam of an mage, f(x, y) no fou nal sub-anges, as follows: Range Range = ( Range mn f cp, 4 3 = In ou poposed opmzed choose he checkpon wh = cp,, cp b) Phase. d) Phase 3. c) Hsogam. cp, cp 3 ) fo whch we as he nal esmaed heshold T n. Nex, we wll naow he seach ange fo he nex sage o one of he wo nal sub-anges bounded by hs checkpon (nal esmaed heshold T n ). To selec he sub- ange, we wll do he followng:. Check CV fo he wo gay levels aound (T n -, j= cp = jp j jp j j (x, y)),, cp Range = cp,, cp 3 3, max ( f ( x, y )) seach echnques, we wll he maxmum ( ) and se (9) (0) () () (3) (4) (5)

4 430 The Inenaonal Aab Jounal of Infomaon Technology, Vol. 3, No. 4, July 06 T n +) and compae o CV a hs checkpon.. If max CV s a (T n -) we selec he sub-ange ha ends a hs checkpon. 3. If (T n +) has max CV, we selec he sub-ange ha sas a hs checkpon. 4. On he ohe hand, f T n self has max CV, we hen wll sop ou seach pocess and se he opmal heshold o T n. Now, n he followng sages and n case we ddn ex he seach pocess, we wll se ou checkpons dffeenly. Fo he cuen sub-ange R c =[, ], he hee check pons (cp, cp, cp 3 ) ae se as follows: ( ) cp = + cp cp = + cp = cp + 3 cp These check pons wll agan dvde he cuen hsogam ange no fou new sub-anges, as follows: Range =,, cp Range = cp,, cp Range = cp,, cp 3 3 Range = cp,, 4 3 Once agan, we wll choose he checkpon wh he maxmum ( ), hen we naow he seach ange n (6) (7) (8) (9) (30) (3) (3) he same way done n he nal sage; Ths pocess wll keep epeaed unl he opmal heshold s found o he seleced sub-ange s small enough ha s when he ange s made of only 3 gay levels. The poposed seach algohm can be summazed n he followng seps:. Compue mage hsogam hs(f(x, y)) and pobables of each nensy level Pop(f(x, y)).. Fnd cp, P and cp 3 usng Equaons 9, 0, especvely. 3. Compue ( (cp ), Equaon Fnd Max j= :3 ( (cp j )). (cp ), (cp 3 )), usng 5. Se nal heshold T n equals o seleced checkpon cp j. 6. Usng one of he anges defned n Equaons, 3, 4, 5 se nex seach ange R c =[, ] as follows: a. Check CV fo he wo gay levels aound (T n -, T n +) and compae o CV a hs checkpon. b. If max CV s a (T n -) we selec he sub-ange ha ends a hs checkpon. c. If has max CV, we selec he sub-ange ha sas a hs checkpon. d. On he ohe hand, f T n has max CV, hen opmal heshold * s found and se o T n and we sop ou seach pocess. 7. Whle (sze(sub-ange)> 3) and (opmal heshold s no found) do: a. Fnd P, cp and cp 3 usng Equaons 6, 7, 8 especvely. b. Compue ( (cp ), (cp ), (cp 3 )), usng Equaon 7. c. Max j= :3 ( (cp j )). d. Usng one of he anges defned n Equaons 9, 30, 3 and 3, we check and se nex seach ange R c =[, ] n he same manne descbed n sep If sze(sub-ange)= 3, hen cuen sub-ange R c, Compue ( ), usng Equaon Se opmal heshold * o wh Max ( ). Table compaes he pseudo code and pefomance fo ognal Osu and poposed Osu-Checkpons on a sample es mage. Table. Pseudo code and sample esul fo ognal osu and osu-checkpons mehods on mage (ensen). Osu Seps Oucomes Se Range o all Gay evels 0-55 Fo all n Range: {Calculae CV} cv(0), cv(),..., cv(55) Fnd heshold max wh max CV max= 9 Poposed Mehod: Osu-Checkpons Seps Phase (Oucomes) Phase (Oucomes) Phase 3 (Oucomes) Se Seach Range Fnd checkpons (cp, cp, cp3) 77, 09, 34 85, 93, 0 87, 89, 9 Defne Sub-Ranges Fomed by Checkpon [0-77], [77-09] [09,34], [34,55] [77-85], [85-93] [93-0], [0,09] [85-87], [87-89] [89-9], [9-93] Calculae CV fo 3 Checkpons CV(77), CV(09), CV(34) CV(85), CV(93), CV(0) CV(87), CV(89), CV(9) Fnd cp wh max CV Cp(77) Cp(93) Cp3(9) Check cv aound seleced cp(cp-, cp, cp+) CV(76), CV(77), CV(78) CV(9), CV(93), cv(94) CV(90), CV(9), CV(9) If cv a cp s Geae han CV a (cp-, cp+) Then ex loop. No No Yes, Ex oop max = 9 Else Se New Range o he Sub-Range ha Cove Value wh max cv Max cv a (78) Max cv a (9) New-Range [77, 09] New-Range [85-93] --- If new ange s same as nal,then ex loop No No --- Else Repea Pocess Repea Pocess Repea Pocess --- Compang Pefomance Vaable/Mehod Osu Poposed Numbe of Gay evels Checked (Ieaons): 56 5 Compuaonal Complexy: O(56 ) O(5 )

5 A Novel Fas Osu Dgal Image Segmenaon Mehod Expemenal Resuls The poposed mehod has been mplemened and evaluaed usng 80 eal mages wh dffeen dmensons and gay levels. Those es mages ae fom Gonzalez [4], whch ae wdely used n eseach papes elaed o mage segmenaon wh dffeen ypes of mehods and echnques. Fs he ognal Osu mehod (Osu) s appled on hese 80 es mages; hen he same mages ae segmened usng ou poposed fas Osu mehod (Osu-checkpons) and he esuls ae shown n Tables and 3. Fo evaluang he pefomance of he poposed mehod agans ognal Osu, we have used hee measues: Numbe of eaons, compuaonal complexy and he esmaon of accuacy. Wh numbe of eaons, we efe o he sze of gay level ange, whch has been esed fo maxmum beween class vaance. In he convenonal Osu, hs s equal o he full gay level ange. The second measue s he compuaonal complexy and s elaed o he numbe of eaons calculang s ( ). Consdeng he me complexy noaon [3], n he convenonal Osu mehod, fohe full ange of gay levels [0,..., ], eques O( ) cycles o calculae all s ( ) []. In ou opmzed Osu mehod, we do no s ( ) calculae fo he full ange of gay levels bu ahe hs s done fo a subse of gay levels. The sze of hs subse, denoed as s, s no fxed and n he wos case wll be s< /3, hus he compuaonal complexy wll be O(s ). We can clealy see ha he compuaonal complexy s elaed o he sze of he esed gay levels, whch s equvalen o numbe of eaons. Fo he hd measue, we have compued he accuacy of ou mehod by esmang he heshold values by compang he esmaed heshold o ha found by he ognal Osu and hen compue he pecenage of exac maches. We have evaluaed he pefomance of ou mehod n enhancng he convenonal Osu by he pecenages of educon n eaons and compuaonal complexy. The pecenage of compuaonal complexy s compued based on he assumpon ha fo each he s ( ) compuaonal effo fo calculang s bounded by c, whee c s consan. Snce, c s a consan fo any value and he compuaonal effo s affeced by he numbe of eaons (value of ), hen he pecenage of educon can be calculaed as: n. ( s ) n = (33) Whee n s he numbe of es mages. As shown n Tables and 3 (Appendx ), ou mehod yelds excellen esuls, hs s demonsaed by he fac ha we succeeded n esmang he same heshold value found by he ognal Osu mehod wh 00% esmaon accuacy, whle educng he compuaonal complexy by an aveage of 99.0% and wh 90.83% aveage educon of he numbe of seach eaons as shown n Table 3. As descbed eale, he opmal heshold value does no always le aound he global mean; hs can be clealy seen n Tables and 3. Fo example, n 5% of he mages he opmal heshold value does no occu n he aea ceneed aound he global mean. Ou poposed echnque has succeeded n selecng he coec aea whn whch he exac heshold value s esmaed by Osu mehod. Theeofoe, he esuls clealy show ha ou poposed mehod yelds a hgh accuacy of heshold esmaon wh a sgnfcan educon of he compuaonal complexy when compaed o ognal Osu counepa. 6. Conclusons Dgal mage segmenaon s an essenal ask n many dgal mage-pocessng applcaons. Obanng bee and moe effcen mage segmenaon s a ccal ssue n hese applcaons whle sll educng he compuaonal complexy. Ths pape has poposed a fas Osu hesholdng echnque. Fom he evaluaon of he accuacy of heshold esmaon, he educon of he compuaonal complexy, can be concluded ha ou poposed mehod (Osu-checkpons) s able o poduce he same heshold value compaed o he ognal Osu mehod bu wh a sgnfcan educon of he compuaonal complexy. Thus, makng useful n many applcaons whee eal (nea) eal me s equed. Refeences [] Alsaeed D., oudane A., ElZaa A., and Sammouda R., Two Modfed Osu Image Segmenaon Mehods ased on ognomal and Gamma Dsbuon Models, n Poceedngs of Inenaonal Confeence on Infomaon Technology and e-sevces (ICITeS), Sousse, pp. -5, 0. [] azen. and Geez S., Segmenaon of Fngepn Images, avalable a: hp://ceseex.s.psu.edu/vewdoc/download?do = &ep=ep&ype=pdf, las vsed 00. [3] Chen., Guo., and Du N., Mul-evel Image Thesholdng. ased on Hsogam Vong., n Poceedngs of he nd Inenaonal Congess on Image and Sgnal Pocessng, Tanjn, pp. -5, 009. [4] Chen Y. and Chen O., Image Segmenaon Mehod Usng Thesholds Auomacally Deemned Fom Pcue Conens, EURASIP Jounal on Image and Vdeo Pocessng, vol. 009, no., pp. -5, 009. [5] Chung K. and Tsa C., Fas Incemenal Algohm Fo Speedng Up The Compuaon Of

6 43 The Inenaonal Aab Jounal of Infomaon Technology, Vol. 3, No. 4, July 06 nazaon, Appled Mahemacs and Compuaon, vol., no., pp , 009. [6] Comen T., eseson C., Rves R., and Sen C., Inoducon o Algohms, MIT Pess, 009. [7] Gonzalez R. and Woods R., Dgal Image Pocessng, Pence Hall, 00. [8] Z., u G., Cheng Y., Yang X., and Zhao C., A Novel Sascal Image Thesholdng Mehod, AEU-Inenaonal Jounal of Eleconcs and Communcaons, vol. 64, no., pp , 00. [9] Mohdeen F., Namah M., and Seenvasagam V., Compason Of Segmenaon Algohms y a Mahemacal Model fo Resolvng Islands and Gulfs n Nucle of Cevcal Cell Images, The Inenaonal Aab Jounal of Infomaon Technology, vol., no. 5, pp , 05. [0] Naayana C., Reddy E., and Pasad M., Acle: Auomac Image Segmenaon Usng Ula Fuzzness, Inenaonal Jounal of Compue Applcaons, vol. 49, no., pp. 6-3, 0. [] O Goman., Expemenal Compasons of nazaon and Mulhesholdng Mehods on Documen Images, n Poceedngs of he h IAPR Inenaonal Confeence on Paen Recognon, Confeence : Compue Vson and Image Pocessng, Jeusalem, pp , 994. [] Osu N., A Theshold Selecon Mehod fom Gay-level Hsogams, IEEE Tansacons on Sysems, Man and Cybenecs, vol. 9, no., pp. 6-66, 979. [3] Pham D., Xu C., and Pnce J., A Suvey of Cuen Mehods n Medcal Image Segmenaon, n Poceedngs of Annual Revew of omedcal Engneeng, pp , 000. [4] Rau S., Raghuvansh M., Dhaaska R., and Rau A., Image Segmenaon A Sae-Of-A Suvey fo Pedcon, n Poceedngs of Inenaonal Confeence on Advanced Compue Conol, Sngapoe, pp , 009. [5] Sezgn M. and Sanku., Suvey Ove Image Thesholdng Technques and Quanave Pefomance Evaluaon, Jounal of Eleconc Imagng, vol. 3, no., pp , 004. [6] Wan Y., Wang J., Sun X., and Hao M., A Modfed Osu Image Segmen Mehod ased On he Raylegh Dsbuon, n Poceedngs of he 3 d IEEE Inenaonal Confeence on Compue Scence and Infomaon Technology, Chengdu, pp. 8-85, 00. [7] Xu., Jackowsk M., Goshasby A., Roseman D., nes S., Yu C., Dhawan A., and Hunley A., Segmenaon of Skn Cance Images, Image and Vson Compung, vol. 7, no., pp , 999. [8] Zhang J. and Hu J., Image Segmenaon ased On d Osu Mehod wh Hsogam Analyss, n Poceedngs of Inenaonal Confeence on Compue Scence and Sofwae Engneeng, pp , 008. Duaa Alsaeed eceved a CS degee n compue scence n 99 and an M.Phl. degee n compue scence (mage pocessng) n 005; boh degees fom he College of Compue and Infomaon Scences (CCIS), Kng Saud Unvesy, Saud Aaba. She s now sudyng fo a PhD n compue scence (mage pocessng) n he Depamen of Compue Scence and Dgal Technologes a Nohumba Unvesy, UK. Fom 993 o 005, she woked as a eache of compue couses fo hgh school sudens wh heang dsables n he Mnsy of Educaon, Saud Aaba. Fom she woked as a geneal educaonal supevso fo specal educaon n he Mnsy of Educaon, Saud Aaba. In 008, she woked as a geneal echncal and educaonal supevso n he Techncal and Vocaonal Tanng Copoaon; n 009, she joned Kng Saud Unvesy as a lecue n he College of Compue and Infomaon Scences (CCIS), and he eseach neess ae n mage pocessng and magesegmenaon. Ahmed oudane eceved an Ingeneud Ea degee n eleconcs fom Ecole Naonale Polyechnque of Alges (ENPA), Algea, n 98, an M.Phl. degee n eleccal engneeng (VSI desgn fo sgnal pocessng) fom he Unvesy of Newcasle-Upon-Tyne, U.K., n 988, and an Ph.D. degee n eleccal engneeng (compue vson) fom he Unvesy of Nongham, U.K., n 99. Fom 99 o 994, he woked as a Reseach Develope n elesuvellance and access conol applcaons. In 994, he joned Queen s Unvesy elfas, elfas, U.K., nally as ecue n compue achecue and mage pocessng and lae on he was pomoed o Reade n Compue Scence. He s now a full Pofesso n Image Engneeng and Secuy and leads he Compue and Eleconc Secuy Sysems Goup a Nohumba Unvesy a Newcasle (UK), and hs eseach neess ae n magng fo foenscs and secuy, bomecs, homeland secuy, mage/vdeo waemakng, cypogaphy and moble and vsual compung. He has auhoed and co-auhoed moe han 50 publcaons and one eseach book. he s a Seno Membe of IEEE.

7 A Novel Fas Osu Dgal Image Segmenaon Mehod 433 Al El-Zaa was a seno sofwae develope a Depamen of Reseach and Developmen, Semconduco Insgh, Oawa, Canada dung Fom 00 o 004, he was an asssan pofesso a he Depamen of omedcal Technology, College of Appled Medcal Scences. Fom 004 o Apl 00, he was an asssan pofesso a he Depamen of Compue Scence, College of compue and nfomaon Scences. Snce Apl 00, he s an assocae pofesso a he same depamen. Pesenly he s an assocae pofesso a he eu Aab Unvesy. He has publshed numeous acles and poceedngs n he aeas of mage pocessng, emoe sensng, and compue vson. He eceved a Sc n compue scence fom he Unvesy of ebanon; eu, ebanon n 990, M.Sc. degee n compue scence fom he Unvesy of Shebooke, Shebooke, Canada n 996, and Ph.D. degee n compue scence fom he Unvesy of Shebooke, Shebooke, Canada n 00. Hs eseach neess nclude mage pocessng, paen ecognon, emoe sensng, and compue vson.

8 344 The Inenaonal Aab Jounal of Infomaon Technology, Vol. 3, No. 4, July 06 Appendx Table. Esmaed hesholds and evaluaon fo mehods: Osu, poposed osu-checkpons (pa-). Image Theshold Global Checkpons Seleced No. of % % Osu Poposed Mean Cp Cp Cp 3 Cp Is Ied Cmplxed Chckenfle wh ones Ensen MRI of Knee Unv Mch MRI Spne Vandy Skn Cance Skn Cance Skn Cance Skn Cance Skn Cance Skn Cance Washngondc and acea lob Ognal lobs lobs In Ccula Aangemen oas an an Tomu () an Tomu () an Tomu (3) an Tomu (4) ubbles uldng Ognal Cameaman Ches Xay Vandy Columba Cabpulsa Opcal Cskull Cygnusloop Xay Ognal Dak lobs on gh ackgound Defecve Weld Denal Xay Face Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Fb Headc Vandy House Cond. In Tabel 3 Table 3. Esmaed hesholds and evaluaon fo mehods: Osu, poposed osu-checkpons (pa-). Image Theshold Global Checkpons Seleced No. of % % Osu Poposed Mean Cp Cp Cp 3 Cp Is IRed Cmplxed Kdney age Sepagon ef Hand Xay ena ung Nosy Regon Odeed Maches Og Ches Xay Polymesomes Rada Rada Rada Random Maches Rce Image wh Inensy Gaden Scalp Skull Small lobs Ognal Thd fom Top Tooh Tungsen Flamen Shaded Tungsen Ognal Tubne lade lack Do Weld Ognal Wood Dowels Yeas USC Evaluaon of Pefomance Among all Expemens (Tables, ) Fo Poposed Mehod (Osu-Checkpons) Accuacy = 00% Pecenage of Reducon n:. No. Of Checked Gay Values = 90.83%. Compuaonal Complexy = 99.0%