_ ONIT -V IMAGit SEGIMENTATION Image seqmenlation has crneaqedl a ar ivepesxtant phase in image-based applications Seqmentalion is the process of Partitioning. a digital wage into nulliPle -gegions and extracting a Heaning-full vegion Kroon as the vegion of inlevest (Ror). Regions of interest vary joorlh applications Detection of vison muilies the thee basic types of quay level discontinuities iin ckigidal irnage ate the ellocirg. | | Poinls 4. Lines 3. £elges An téoloted point is a pommt whose gray level ts backqxourcl ma homage- 2 _ Significardly, diffesent foo ils neous anca. A genexic 3x3 spatial yn mask 48 Shown in below fig (a+ z leolza | [= [zz [2s] 74 | 2516 | 11 | % | 29 spatiol Mask Fig(a) * geneaic 3x3 I ee onik-& | 1153Se ee \| she Mask jo Supey wmposed orto an wage and the lGonvelution process 18 applied. the vesporse of the Mask 7 iS given as, R= 2 2h, \| where , the f, Values axe the gray level voluies of the | Pixels associated with the image - A dhyeshold value T is used to . dendify» the Me poinls. a Point is said i 4 : | to be detected at the location on which | ( [4 | the Mask 1 Céntyed if IRIZTs . = Mask volue of where 'T? 45 a won —negative inteqea’ The a point detection Mask al points. @)line petection:— ae ee x the next level of complerity 39" Line detection” * G@nsider, the Masks Shown below. these axe used fon the line detection. ; : i : be Fh |! 2) Jaleeel [ey--| > jee eee ee ler Q | [-1[2 -I il a|-w| | - : {al -1Pa| ee eee ee | Hoxiaandal + 45° vertical 45° veike & ¢ BI3 ‘ett the fivst ae ee = a, i were moved atound an ima! il | image, 7 | would yespond mone shrorgly» do fin ay | hosigentally mes garerde eee na back quourd , the Max. xspanse wculd | | when the line passed thyough the vdle 3002 = 7 Mosk: his 19 easily veaifted by stelchings a evey of 13 wilh a line of 0 deffexent 93a4- | eee (say, 5's) semningy hoyiaentally bhvougyh 2 ng # ar mask yesponds best 4o lines | | ontented at 45°, the 94 ask fo vertical lines and the 4th mask to lines } "ys? dixection. “These divections can be established also | | the pueffcved direction of cach mask coefficient City 2) than othe possible discetions | a bager «wile that, the o- efficients yesponse by noting’ tak | 9 weighted osth | in each’ Mask stm £0 re, | | mndicalingy 4 7020 fsorm the Masks in anttas 2 of a@nstant gray’ : ard Ry denote left to Right | fae Yun individually Jevel - | ied lef Ry Ra oR Lhe sesponses of the Masks to above — figure , form ‘x suppose that the four Mask? {hyough 9” gmage an the image ’ IR] IRL ysl, a pat a cextain point } es gid fo be More - likely associated with | q fne in Ihe disection of Mask ‘1’.Ext- rh at a pond tn the tage, IR) 7 IR;) fox point 15 said fo be p22.3,4 that particular wilh a besizental line more likely associated interested in detecting all the [ines in *% we ate defined by a diver Mask, an tmage 'M’ the direction Lhsoegh the mage ard bhicshely we simply run the Mask the absotuke value of the aesull RX the points that are lef we which , for lines one Pixel lo the dixection defined by An “Edge” +5 a set of Connected prwels that lic on the boundary bhy two vegions image, an édge 15 a lecal fealuse two yegjons im the strongest 2esPonses Lhick » correspond closesé fhe Hask- e Tha qxay~ scale that | within a neighbouthood , separates each of cshich the guay - level 45 mose (ox) Jess uniform with different values or the two sides of the edge. & dioilal €dee nodal of an ideax the lhickness of the edge % delexmined py the length of the somp, as it faansilions fyorn an dinilal length 45 delesnined Vy The hy, dhe degree of to a firal qoay-level. the slope. which a 4rtann , $5 delexmined blwecing # Blovred elges tend to be thick and shasp eclges tere fo be thn de the below figure (a) shows the image from eahich figtb) qwas erdyacted - = ty th) shows a hositontal qyay-level paofile of dhe® edge bo the {wo yegions. # the Magnitude of the fiast devivedive * presence Grway-level Can be used bo detect the of an edge af a poimt tn an image profile : (ee) do determine fa Point 16 07 : ‘frase ae dexivol ve a yams # [lay , the stan of the Second dexivodive | fan be used to delearnine » whethe x | =| | an edge piel ties in the clavk (03) t fight side of an edge oa \ the 2 puopentics of the second devivetive ayound gn ecge- Jwo values for © tn an tage i) rh produces vey” ae | fan ures sible Seatwres j- and |i An jag inet line and “ve! values yy straight goin the a extaeme ‘pve’ of the second de yivative ovuld cxoss | | nite & ) S1530b zero viecay the mid-point of the edge» This poopesty | $s known as "1010 - Cx098inq propeat”- qhis paopesty 15 cuiolely used for focaling the conters of thick edges edge defination 19 dhe most Common Lechnic for definatingy discontinaties in infensity values the vawious edge lefenalions available in function edge ane @ follows Is Robexts edge detector , pavadient ope vatos S 2. paewill's edge detecto* 3. gobe] edge detector’ - 4. Laplacian of a Gaussian edge detector 5. 7x0 - Grossing» detector €+ Canny -edge delecloy (0 Swaclient apezes | fissk -oxdex derivatives of a digital image o%¢ based (on yevctous approximations of the ap gaadient- the | grodicnd of an tmage frog) ak location Oy) 15 defined — gym | as the vector f vt = [e] : » ony we [70 fy * for edge defection, we have fo consider the Magnitude of this vector ,deroted vt ), where, Ne] ) af-af-| salolt | | pe | o lo | a] ol! i | vpn alold | fox # fos 2 35 (* In this formation, the diff bay the fixst and Lhivd || | Yows of the 3x3 image vxegion appro xemates the | deyivative in the x-divection (y-dtyection ) \* the prewilt delector +3 Some whal easjey to execute | the sobew detector . bat A genesate| | | Gmpatationally an I | shghthy neisey veselts | ape al Jobe! edge detector: - | the below figene shows dhe masks used by a sober elge detectow to estimate bx and Sty | } | Gx = (274 2rg t tq) ~ (% ARWet%3)— (9) | | Gy = (tat Q264tq) ~ (2 @2y 427) CD) || - | x weight value of (2) i used 0 achieve Some Srool bing | by givinge move importance to the centex powd | ton uy, 3153Prewdl — Sebo (3) Laplacian opexators :- ORAS RT 3 | yotation the laplacian of a a-p function flxy) 13 a second oxdex derivative is defined as vs: a 3, lo) for a 3x8. vegion, one of the devo fowrns enceunteaed [Host facguently, in practice 15, VS = 495s— Grazys 47g) A digital appacuimation sneluding, the diagonal neighbours 75 aver by, vi = 8t5- Gar gap te 4tur? ant Shown heae 121478 +2qg) 22 my these #00 for implement eigns @ (R) ant * Masks ss0taoPic the amplementation of the ancuernerts of ag ard us respectively.reer ernnaaeeaaeeeeeeeraa the laplacton gene vlly- | form — fox" edge detection | she laplacian 49 Combined wi Precursor 40 finding, edges via ze vo - CxOSIMGS Gnsider the fanction | 45 nok used in its 0%54 nal | seveval yeasore th Smoothiny 05 a for Povten ata | where yy" ev4y* and be" is slandostd dewiation | # Grvolvingy this fn with an irnage Huts dhe image, with | | the deqeee of bluxcing, beiny determined by dhe vake | oF fe’. | | the (laplacian of bn! 15 GG SMe Yho) bx dy | whe) = PT ra] -Vas* | ot é ty) | | qhis function 18 commonly de reffered to as | } laplacian of a Gaussian (Loc ) because ¢9'n (3) 18 im the = | form of Graussian f'n. | | tthe below figuve shows cxoss - section of the log | fin. gue fo ls shape » the lag, sometimes is cated | | the Mexican tat function Unie WlS3a) Cv035- Section showings ~3 2ex0 C¥0ssings 4) zeae - Cuossing, dedecto2'- RA AA ann This delecloy 1 based on the same ide othe diffesence 15, in this delectox , the log, detector Convolution 5) Canny edge delectos:- This is the Most pouesful od 15 pexfosmed usingy a specified filles SH oO |-! =a ate fuqea| wf-a| | | | po jot {als o | fofe lle fe | b) 5x5 Mask appacximation fo the shope of the 39 image cha a as the et ge detectors supplied => by the fn edge the procedune employed bp Commyy edge detecton 1 cullined below - 1) the trage tg smothed using» Goussion files cwith a specific 40 seduce noise. ii) the Local distance qyadient 9 0uy)= Te reat andl edge dtveclion 0047= tai" (ZE ) ate evaluated af cach poinout "Gobet" ov © froyett* % Robods’ onit-t 531a deleclos Can be used to calculate Gry’ ane ony? | edge be a poinl whose Strxerqlh | An edge point 43 defined fo [1 beallyy Max. tn the discelion of graudi | tii) the edge point delesmined in Step ci? give vise to Vay tn the — qradout Magnitude image the algosthm then racks along the fop of these xidges and sels & Zevo all pixels © that ane not actually on the widge top S as fo gue a thin tne in the op, a process known 9 " Non - Maximal Suppsession’ = The yidge pixels ane then thresholded using, foso Lhyesholds T and T2 with T«T2 - Ridge Pixels with values qreatey than 't,'. ane gaid to be strong, Cdge pixels Ridge Pixels with values blw Ty and Said fo the « weak" edge Prvels. Finally, the algorithm performs edge hindhinge by- ‘incoxpanting- the weak Pixels ane $-Cormecked to a = Strong, pixels. bnik-4, 12 [53_ Edge linking? and boundosy» eet eee fe crrtacae wee Fdge linking, procedure ts used to place edge Pyxels into meaning fall edges. the different Methods for edge linking, axe follows 1) local processing, 1) Global processing, vio the Hough sansfoam | tii) Global processing» via the qaaph thesomlic lechnigucs = 7? Local Paocessing :- Jn lscal processinge the edge linking» cawvicd | by, foo methods othe stxength (wognetude ) of quadient opexator to produce edge pixel: 2. Gradient vector dxeation co-osdtvoles of cdge Pixcb 05 C*or¥o) ig Used Qansider the 2 and the co-oxdinates of high - pixel 05 (oY) the Magnitude of 6dge pixel and nighloowred Pte] in? whe ny Ivf Coy) - Cot Gordo) [EE wheve, € vepsesen’s von - negative threshold the yangk of edge Pixel and nghbouned Pre! Same — when onit-&, 13/53TR (x. W~ 2 C140) [ZA where, A> yen -negative angle threshold # the direction of fhe edge af Gey) nigh ts Ped fo the pixel at Coy) tt both bounced of (ry) Magnilude ard dixection criteria bat wte satisfied . this process 15 vepeated every location in image Global processing» via the Hough Tan stoxms :- Tn this processing» when the points lie on a specified shape then they ane linked fdge [inking» binany tmage detexmine gradient of * To. obtain binany rage delesmine, thexs Qaodient of image * specify, sub dwision tn the §-6 Plane ® Detfearnine the Counts of the acculurmotons Pixels connectedions. * Determine the velalionship method chosed. ~ Suppose that , fry « swe ”’ ponls +5 an find subsets of these points . line 7 ONE apassible 2 Sln fist deteym: ne of Points — that axé chse to Pasticulasily, Problem with this procedune iS then . ip bnik thers hold and hold anc calls x high Considey global image , we want to thot lie on . straight al( subsets lines - the involves 4) ruls3finding: ne ym fines and then pevformning, ony Grn-V 2) rw? Compaising, - of cvesy Point of bres. This appacach 18 Computational pyohibidive ‘yn all by! the Most lyjval applications . Hough pxopred on altenative appsoach , commonly yeffexcd fo as Hough txansfoam. ; Gnsidey a point G:y:7 and the gereval en of a 7 staaight ive the slope - itercert fworm Y: sax +b for . vatying, values of a and ‘b’ ynfinety lines Pass thaough (oy) but they all satisfies Year -+b fox diffexent ‘a ard 'b’ values y ‘ a a) A d ay NN os) \ N\ “contameley space (ob plre? xy Plane + However, writing this ein as be-%Gat¥; ard covvesP- onding the ab plane yields the eqn of a single [ine for a fixed fair C94) = ontt-& , (5153“ has a line in fasameter ¥ A Second poink Oxy -%) also Sand this ling interscets Space associated with ih the line associated cwith (14,45) and (a! b'), wheac ‘a? snteacept of the line ond 41 43 the slope ard xy plane Gontaining » both Ge oYy) and Os 94:) in the ¥ Trfact » all pomls contained on this line have lines tndessect at (a’,b) which Mm parametex Space thal axe shown m above figure. * The comudational allsactiveness asses fyorn Sub deviding- the poxametey Space jnlo So-called “accumulator cell's ,ag given in below figute » wheat (rox Gn) Grd Cbyrax ybmin) ate the expected sanges of Shope and intercept values of the Hough fxansfoxm, «The cells at co-oadinates C3), estth accumulates value A Cis) cossesponds dren ee i | Sub division of I Porametey plane je Ee ++] for use in the le | Hough dyansfex | el Lae | omy, ILS e| | ¢ Inilially these cells are set do zexo. # Then» for VERY oink Cn Ye) in. the image plane, we h metey tq? kt the pavameley ‘ equal each of the allowed sub division values on the a-axis and salve for the Covyesponcling ‘y using, the eqn bs -Xqa4 Yee she vesulling b's axe then rounded off to the veavest allowed value tn dhe b-aus. # Ta choice of « , A eb ‘ap'. yegulls in solution by , we FC A (p9) = Ala) 41 ¥ AE the end of this’ procedeote a value of ‘a? in Alii) | Cossespords "a" pols in dhe ayeplane [ying on the bine Yrax +bi- # Note that » sub dividing, the an a-axis into ¢k’ incacm énls give S$. foy every pomt Cx 4k) ok values of ‘4! G@xyespording, fo the k- possible values of fq’. ¥€ with 'y’ sage poinls this Method involves mk 3 | Computations | A problari with using» the e@'n Y=axtb to yepyesent a tine 15 that , the slope -appacaches infinity as the line approaches » the vertical one way around this difficulty 1s Jo use the vesmal yepresentation of a line. onit-4, les+ | IN ‘ | 0 \ Yy J. woxmat | “ representation [ of a line ¥ # This cnslaucting, a fable is Sams 08 fox slope - intcacept representation. ; He % Instead of straight fines , however the locci we | gmusoidal cunves in the 36 Plone. * Incrementirg, "> and solvingy for the coxsesponding, 'p’ gives '@? entiaes in aceurulator A(ns) associated with cell detexmined bi (5;.64)- the above figute shows the subdivision of Pasametey Sface. hese , the yange of angle 'e? 5 $40, weasuaed W-a-to y-axis . = thus fromm Figcay ja horigertal- ne has @=0°, wilh ‘g) beiengr egual 49 tve ‘X-intexcept.. “# lly, a vestical line has 6-90» with ¢s being egual | fo the tve y-intercert 09 @=-qe° with '4! fe | egal to the -ve y-imtexcepe . i Ralthough , the focus So far hos been on staaighh kines, | the Hough transform is applicable ing to any, function of oNnit-q) teesfe forra give) =e, whee fv! 16 a weeloa" of Go-ordlinales and '4' 39 a veckor of @ efficients. 0-4 (Yay = a” 30) Fox’ example , the pomls lying or the civcler Can he dlelected by using, the above app3aach x The basic difference is the process pxcsence of 4,2 ma 30 pavametey Space with ceurnutators of the form ACW Solve the ° and ¢ which vesulls cube tke Cells and a il the proceduse 15 fo gmexement cy for c what Uhat salisfies e9’n ¢2) and accurnulator Covyesporingy to the cell [he tviplet (e162 1) she Complexity” of the Hough tyan on the no-of Co-oxdinates and ¢a> update associated otth sforrm 16 Sdzonqly “dependent - and ¢o-efficients im a given Hexe 1 the €dge linking, 18 based on sepsesenting 15 in the form of o graph and Scaaching. that coxyesPorid Be functional zepaesentation. edge seqment © Lhe qzaph fox low cost paths significant edges - # This yepresertation pexforms well in the Pacsence of noise - nit = | tals 315 Gnsidexably more Complicated This procedure than the other more processing lime and veguircs methods pefine on edge clement as the boundary bho two bhok “Pp! and tg) ate 4 -nicighbeurs pixels tp’ and '9! such as figuné here shows #kdge clermends axe identified by the xy - co caclinakes of pomls ‘Pp’ ard 9) te, m dhis figuae » Lhe edge clement 43 defined by dhe Paivs Crp. 4p) Cxg,¥g) 3x3 ‘frnage Shown below Consider the # the ouley members one pixel co-ovdinales and tae numbeas tn baackels yepyesent qyay-level value. * bach edge element , define d by pixels ‘P' anal tg! has ah associated cst ,defined as cCp9g) = H- [tcp) - 99] where , H-> Highest gaay-level value in the image FP) % FQ) 5 Gaay level values of © p? and te! vespectively wall: | | 90/53¥ By? e@nvention tha Ponk tpt ay on the RRS af the direction of travel abong. edge elernents- pe qhe edge Seqment tay (aa) 46 between {he poinks (hay ane (229 tm the above fig Cb) ers we Pavelling, to the lef bios the $a hen 'P2 49 Chay ancl q 48 (22). Tr Fhis Cose cost 8 'g) as shown abeve the edge Segment in fig tb) ime [oo poinks, We assume that , the adges start in the top sy vow and terminate tn the last vou , 50 that the ast clement of an edge can be only» plo points (1) 02) oo (12) 9 Ch3)- | le Wh , the last edge element has to be blo the points (3,)) G2) (ort) (3,20 1 3)3) ¥,* p) and 9g? should be u-neighbouss only. the below figate shows the qxaph fox fhis Psoblam. = a 2 a a D (12) 3 3) aoc a a 6 — feo f a ze a) (12) pees} _B. -- [ora] ry) aS @ a Na ioe = { Fy Lol eae [ene] 3 (2) F229 | onit-y, 253#* Heac y cach rede covacsponds to an edge element from above figuste AN are crisls bis two nectes if the Loo coaxespording, taken in succession Can be past af an edge elements Cdge * Goal yaces ané Shown shaded. : * the Min. cst Path ts shocon dashed ,and the edge fo this path is shawn above in fiqes Gyvesponds Types of Image seqmentation appsoaches’- ET Ra ae tO TN o- 1) pixel based appyoach:- H 45 one of the simplest approach fox Seqmentation, only, gray values of individual pivels ate used. fi) Region oserted Based Appyoach’- io eee Pe epee 8% quouping- the neighbowting, pirels of Similan iolensily, tevels , an image 1s portitionod ina a@nnectod aca. weve , the qvay-Values ane avalyaed in larger aacas Wi) ie - based opproach: fdge detection 15 the appscach used Host facquenlly fox segmenting, tages based on absupl changes tn indensity - A ‘bias’ m dhe side of scqmented obiect can be avoided without using a Complex Afxeshaldings sheme. baik-q, dafss=) 4) thaesholdingy Technique based ov) method of segmentation which is properly tov) populaaly? used, postant ane ynade. Hee . the decissio pixel formation thacshalding » 15 a havi espe ciallys tre applications where speed 49 an Hm factor 5) Moy phological appacach .~ H is also called os “waleyshed segmentation Tht» approach = Gombines several tye allstbules of segmentation Thaesholdeing :- weet suppose that the gray level hislogyarn shown i figca) Covvesponds te an trrage $004) Composed af light obiecls a1 a dayk background . im such a way that ebicet and background fixels have gray levels qouped, jnto Jaa dominant medes — | | lh. all, l,l alll nie G pasls ,| “Global Thaesheldirg: - * [te simplest of all thyeshokling techniques is to pasts | ion the image hislogsam by using» a Single global | [haeshold ‘rv’. the segmentation 15 then accomplished by- Scanning the | | image, pixel by pivel and labeling cach pixel as object ox back gyound, depercting, on whether gyay- le of that pixel is greater ov less than the value of 'r the success of this method depends entixely on hoa well Lhe fistoqyam Can be paatitioncd- | en The following algoasthr Can be used to oblain 7 | | | | | | autornatically - - Select an initial estimate fox ‘7’. 2. Seqment the image using ‘r'. this will Paoduce two qxoups of prrels. Gi > Gnsisting, of all pixels with gaay-level values > 7 | “ond G2 > Gnsisting of all pivels with values <# 3. Compute the avesage guay-level values su, and HM, fox the pixels in yeqions Gy and Ge 4. @mpute a new thaeshold value = kta | 5 Repeat step 2 through 4 untill the diffexence in | Successive ;levations is Smaller than fhe pycelefined | Paxametey to | wey ont, els- . 4 i pasamelex ‘t,' +5 used fo cantool fhe vo- of ilexations in Situations whese speed 45 a7 smpowtant issue: sve. thxesholding:- ppaplive [heesteddieg: Tmaging, factors such as uneven sMturmination hisfogvar that single qbbal , ada pkive thyesh- gn faansform a perfectly seqmentable gan not be — paxtitioned effectivly by o thaeshold- To handle lypes of stluations olding 18 used - kone obvious wa und 15 fo Select a thyeshold ‘7! y to extzoct the ebiects from the ) that serazates back 9x0 [hese Modes then any point plow for which $004) 77 19 called an ‘obieck_ goin!’ otherwise the poml % called“ backqyound wl Se modes chavactesiac the image * Tn fig (7,3 dominant higlogaar # Herve, Mallilevel belonging to one object fass fh foO4)uTe thresholding, classifies @ point (19) 05 object class if Tr 2(uy) £72, to the and 4o the backq ound other fh fowetr Tn general, seqmmentation pyoblams yequiaing~ Multiple fhesholds we best solved using vegion g vowing Nethods onit-4 85/532% * the hvesholding- may be viewed @S an operation | Ilad invelvs fests against a function ‘7’ of the form | T= Tloy, poow), Foy y | where, fog) > Gray level of point Oo¥) | PCx,yJ—> Some local Pxopeaty of this point Guy) Ex!- Average quay level of a neighbowt hood Centered on CY) A thyeshod image gtey) is defined as = 96,4) = tf fy 2T of fewer — @. * when “> ‘depends *only on f(xy) . the thyeshold 45 galled qlebal — thyesholding . *H "Ty depends on both $004) and per, the Ubeshebtiny is called ‘ focal thaeshalding’. #0 7 deperds on the Spatial Co-oadinates ‘x’ and ty? the fhxeshold is called ‘ om “dynamic? cox) adaplive thyeshaldin 7 f tort Uy) 26/53| tn Uhis approach, the Chacsh | ’ hreshold value fox cach Firel |depercls Upon the locations of Pixels im sub -irnases Lx we have fo ‘devide the oviginal ‘image inte cabs ame then lilac a different fhveshold 40 segrnent cach sub -image all the sub-smages thal did not Contain a boundary blo obieck Containing, boundaaies hod variance’ in excess of loo ; Fach Sub-image with variance gyeatey than [oo #49 | Segmented with a fhyeshold computed fox that gub-irrage using, the atove algorithm ¥ The snifial value of ‘+’ each cose ‘was selected 95 Lhe point mid way bes the Hin and Max. qvay levels in the Sub- image Rall ub -wnages wilh variance less than [000 were fycated a5 one Composite image » hich was Seqmented | using, a single ~ Lhyeshold estimated using the same | algcocithon: | Adaptive thyesholding, 19 impostant in text image! flocs detection and object feadwre | | | binasetgation- | binaaigation » glass onit-& als[cplimal global and ‘adaptive thacsholding:- “ | | this methed establishes Ibyesholds that pycduce | | the Min avexage segmentation éxx0¥ | Thal an tmage contains only wo Pyinespal | | qvay yegions - Let ‘2’ denole qray- level values these woh} | may be xandom w quantities ard theiy histogiam may be Gonsidesed an estimate of bheiy PDF » Plz) | Suppose this overall densily function 48 the Sum ox mintuxe of two dengilies one for the light and the = othey fox the dask yegtons in the image Lhe densities is known oF assumed , | Deleymine an oplimal thyeshold for seqmenting Ue | mage in to the fuo distinct egions. | vo) i | «it lhe form of ih 39 Possible to nye alt 9 i a || | a/ eee Ve | pack gzound « | x this figere shows two ofp's ) Coxsesponds to gray levels of object the backgsound levels. ASSUME 5 Pye jn an image & Pelz) yepyesents unit) asissthe Ni jaduse pay describing» the overall qray-level ywualion wn dhe image is PZ) = Py RIZ) + PaPal2y—>y) where, an - P,~> probability, wilh yalue 2) 43 an obiect piel fa > probability that pixel is background Rexel Given pixel belongs to esther obigct ov backgxound , then P+ Pa zl —5&) AM page 76 Seqmented by classifyny 45 ackgyound all pixels with quay-levels gyeater than a Unveshelel 'z'. All ofhex pyxels axe called obiect pixels our Main absective is to select the value -t1’ that minimises the average evsoy in making the decisions fhat a qiven pixel belongs fo an obiect con to the backgyound- thus , the psobabikty of eysonccusly classifying a backgxound point as an object point is, a Jt E,(1) = { Pylzid2 3G) © his t6 the avea under the cuave of “Patz? to the left of the thaeshold Similonly, the psobability- of exvoncously Classifying, an obsect pot as backgyound 15, co By (D= { p(ad2 WW) + onit- 4, 2als3» which is the atea undley The curve of Pitz) bo the sight of ty! then the overall probabilily of exor 19 ECT) = Pa E\lT) + PEQ(D 25) * note » how fhe quanlilies £, and E2 ave weighted by the probability of ‘occusance of obseck (ox) backgxounc pixels . Hexe subscripis axe also opposites. To find the “ Ihyeshold value fox which this caor 48 Minimal Pi ACD = Pert) 6 find the optimum th qhis egh 19 solved for ‘1’ to Lhseshold. woke Phot +46 fr=pe » then the optimum thyeshed +5 whee the curves for Pitz? & Plz) inteasect. A eA ao 7 (2-2) age Pe) - — lainey + Saree >t) using, thas egin, the ‘geneval sol’n Of cfr G) results oH in the followings soln fox the thveshold T. AT’; BT4¢ 20 6s) whee , = at A= 6*- 6? Be Q (ne*-ug6)) C= atu Suey - yah VEPey In COP CesPed onit-4 36aA quadaalic egy | " has lux possible sal’ny , Lao Jhoeshold values mayb re ’ solulion- yegutaed Lo oblain the aptima) sf fhe vavian ces ay ave equal, 6 6 = ,? et =e? ,a Single lhsesheld ss Sufficient: pe Bu > eae ep Tat 5 200) uf Pi=P2 > : 1 = Pa Lhe optimal {hueshold 1s the 9v9 af : the means ©” © 29 Determining the optimal Ihveshold may be May accomplished for other densities of known fos , such as | Rayleigh and tog -noxmal densities Ex.- the Mean squaxe 2 PB [pees -h(2s)J 30 2 exsoy bl» the mirbuye density +t ems = % > histoqvam “6 assumed ing the complete where an n- Point pal yeason for cotimat the pairs density” qs to deteamine Ihe presence (92) absence dominant sabes i" the POF.optimum Global Thaesholding, using» olsu's Method. RRA RAR RRA Thyesholding- ray be viewed 95 a statistical decision theory, paoblen whose objective 15 ba Minirnige signing. prrcls to tuo o” the avy exxay jncusved im a8 ts known to have move gyoups on classes his problem an elegant closed fovm Sol’n known a5 the Bays decision ule . this approach 38 given above the alfesnalive appyoach fox opbrnurm global . thre. sholding, 38 olsu's Nelhed . the Method 1 OPkimurn tn the Senge Ibat i maximizes the bls - class vaniance uscd m statistical diccsimmal onalysis Tl 15 based enkively on computations perfosmed on the histogyam of an image , an casily, oblainable 4-D annay the algoxithm foy obgu's method 15 a5 follocs *Gompule the noxmaliaed histogram of the i/p image Denote the Components of the histogram by p,, j-o19--t-1 ox * ; Corpede the cumulative sums, pick), Fox koh... b using the ¢€ &N Pick) = £p. ize! eG ; ompute the cumulative sums, mt) , doy keooit . we ouncleny | lo be satisfied amd 2m ] Hi) the Section of Seed points Cte to ZL tegion Pa suppose, the Cals P ne Smooth segions snag posing? acornehrY | | wage Gorse’ fo the planay suxfaces of the See al ahieats | Adtorcingy , one may, assume thal the vasiance tn | | gang level of pitels within a yegion should be Srnall «This | leads to 0 pasticulaay- pxopeaty oy moxe specifically” | “homogeneity, paopearlies fo be sakisficd bi Lhe 4ray-levels | of pixels _ tying. within a zegion | | | lel us denote this pyopeyty, by prop ¢R) whese | 3 Prop (R) = oe fecer -af eth oo | | PO Geader) dialer . | where» Cord (R) 18 the v0-oF pixels in the aeqion ‘r? | ard *q’ is the vneans of qoay-level of prrels lying in that yegion: iC» ~ E9000) 9 * Gard(a) H,0€R and th > pyedefined thyeshold ont gy, 38/53OO le le any ¥eqions would came | ep Th Uh’ is haages mony gegions would mmeige toa single 9egion another problem of [hig Heasuwe of homoge. x4 more poscls neilyr is a5 the 39 se, moxe an jcluded i” Ihe yeqion , the pixels that fave 9xay - tons ave accepted because of £ Yh! be very small, many ti von grows aye | eves with higher vavial overaging « To avoid this problem > yarige of qvay - levels of pixels lying, = we fake the homogeneity. pyopealy for a Small d define the Property 5» Rop (Ry; max $9) J- min S gcc) fe th i moer i J met i ma xegion ani Assigning, one pie! at a time as membex of | the cuxvent vegion 15 Lhe simplest appyoach to region 15 one of the Slowest approach as «ell. | quowing, . This aly is to Gnsidey pn bellex appsoach on the prope anstead of Single Pixel = yegion al a time be salistied , this yeqion. a 5rmall | Dependingr on the propesty to Move robust 40 noise than the Pixel- | basecl appsoach 8 basecl appyoach EX | Consider the tmage shown heve. nk Gy 24/. 2 G lw& | | tT T foy seed pomt 9, sls region wg Ri | Jereesen Sr Se Negea Gab divide an image initially nko a set of cnbitaay sdisiointed aegions and then Merge ard [or | sph lhe wegions to Salisty 10 conditions - | Region sphilling :- \ MA ae let ‘R' yepaesent the entive image segion and select a predicode P. i one approach ox segmenting, R is to subdevide it 1 successively into Smalley and smallax guadsant eqions $0 | That, for anys vegion Ri, PCR) = THE. we stast with enbiae | yegqion - | Ti ptr = false, we divide the tmage into quadants. Tf tp! 4s false for any guadsand » we subdqvide ed ail yon! info subgacdsands , and $ 00° This paaticulos splitting) fechrigue has a Convenient yepaesentation in the foro of a 50 - celled guadkaee - a5 qiven belowTee ts fo a Subdivision I, [ [ex cine yoot of the bree and that ¢ach nade Corvesponds was Gubdivided fusthex | this Gase’ Ry || @ partitioned image b) Coxscspordiny quadtace | = Region Méaqing*- PAA nna tf only, splitting were used 5 the would contain adjacend regions with identical pacpes, yemedied by allowing” Messing, final postition likely- lies. This daaw back may be aS well as splitting Two adjacent vegions Ry and R_ ae Merged only # | i P(R;URK) = Tauc i this 15 sumraviaed by the following procedures 7 (hn, | | which jak any Step we ¥# split into 4 adisiomt quadsants any Region Ry dou which p(R.) = false * Meage any, adjacent segions Rg and Ry foxy which Poy = True | | # Stop when no further Mesqing toy) splitlingy 15 Possible | Tk ® Customany fo specify a min guad yegion 2 | Sige beyond which ne fuathey splitkings 15 Coded cut. | I MER YY S83Region Meaging, 15 exactly opposite bo the vogion splitlinge Methect. this Methed is a bottom -up methed hile vegion splilling ig top -bollom —sacthed ¥* bi i 4 like segion spklbing, this wethod also 1s arpticable te images whose no-of yows and no-of Columns arte same integex poweyx of Q we check if 4 adjacent bemogenecus wegiond aaange! ina axa fashion fogethey saliffies dhe homogeneity prperty TH yes »we merge those aegions fo a single homogencous yeqion, otheawise the yeqions ane left as they ae Ext- Consider an $X3 4mage 14 6 € 55 | 43 Sra te, 2 3 | 4 ¢ a2 | 4? 2 2 | 5 6 fe 3 | 4 23 56 | f # The vegions cy reas Gi 1 Ca axe merged R, because they Satisfy the homogeneity property. Regions Gy Crys Car 2 Coy ~~~ eke ate kept apast - hecause they, individually Satisty, the psopeaty « but vet as a@ a@mbination anit 4 53/34, ; UNIT -W. PART ot OCESSTNGr ] ROREHLORTEAL Hage pROcEssENe tion: | The woud Moxphology commonly, clenotes a bsanch of biolegy, Hak deals with the fowm and sbyuctuye of animals | and plants. we use the Same word hese in the contexd of | Mathermatical Moxphology 45 a tool fox ertaacting. image Cornpen, tris that axe usetul tn dhe vepwesentation and descxiplion of| region shape, such as boundaxies , skeltons vand the Conve y bull. we oe intevested algo in Moxphelogical techniques foy Px - on post Processing, , Such as Moxphologica| Fillering . dhin- "ing, and Pruning . | Exosion and Dilation :- rare nee ADI Feosion:- wet with A and Boas sels im 22 | the exosion of A and by! 8, denoted nog , js defined as, | 498 = f2/(n2cA} —say Th words , this €9'n indicates thak fhe ezoston of A by B is the set of all points z Such that B, translated by 2, is Gntained in A-TIn the following. discussion, sel Bis assumed fo be a Skruckwing, Glemend . £9! a) 4s eon | the Mathmatical formation of the example +n Fig @. | Because the statement that g has to be Contained ina) 1S equivalent fo 6 not Sharing, a Common elernents | with Lhe backqyound ewe Can expyess erosion im theplowing’ equivalent — foyr, ASB = = J2/0B),nae - ¢} 02) hese. aS Ae ‘ Ao is the Cornplernent of A and gis the empl sed 7c follows ‘ the following fiqutes shows of exosion . the clement: of AwB or Shown shaded and the backgsound is while. the Solid boundary in figced 4s the lirik beyond ghich fuxther displacements of the sign of B would couse the 5b vuchuring, an A Thus, the | Gse being, completly, contained points within this boundany 1 Gonstitubes the 1 Cxosion shaxed in figeo | of z that satisty . dashed tn. 3, clement to locus of erosion of A by B. we show th « exosion 13 Simply the set of values egn ) &@ + The boundany- of set A is shown fq. © and fig) only as a ste yefesence vik ts not pal of the €xosion operation . Fig Gd. (e) shows the ¢x0sion of A by gct was extoded ) shows an elongated bfyuctwany clement , and fig a this element - ar note that the oaiginal fo a line. | d CUD | | | ol dry 1 da) | 6 { 7 | oo ; oe 3d/q d/e \ | . ; —|| e Vp dy We Figuae + (aj set A (b) Square staucttoing clement B i (e) Exosion of A by B», Shown shaded | (A) Elongated Stsuctuaing, element (€) Exosion of A by 8 usiny this element with a and B as Selo in 2% , the dilation of A by | B, denoted ADB , 15 defined a : 4 | A@8 = 2 /8),na dd $30 | | this egn is based on xeflectingry B about its oigin, | o I ard shifting, this seflection by z. the dilation of Aly B | | then ts the sek of all displacements 12 1 Such that B ond | A ovevlp by af least one clement . Based on this | inteypaclation » ¢gn t) Gn be wrillen eguivalently as, A@B = $2 /f@),na} ca pa) As before , we assume that B is a ste%ucting> ‘nl 43 the set to be dilated. i | element — and Ul 7 nie yy qe)sa: UZ Fans © and (2) axe nok the only, definilions of aietion curvenllys tn use, Houcver , the paccedingy dent pons have a distinct advantage aver othea formulations wn that they are moze inluttive when the struck” clement 8 15 Viewed a5 a Convolution mask « the baste pyocess of Shipping @ about sls oyigin ond then successi- vely displacing, tk sq that sl shes ; - on) Ides over set (mage) A 4 analogous 6 Spatial convolution . However , the dilation 13 based on Set operations and thescloxe 19 9 von-hneay opeaction, Whereas Convolution 1s a linear opcaation Unlike cxosion , which 15 a Shainking Cox) Lhinningy opevation dilation “'gyows” fox) " Unichness ” objects 1” 0 binastyy image: the specie manney and extent of this thickening 15 Controlled by the shape of the structusiny clement used - figcar shows the Same set used in cxosion fiqutes and figlb) shows a Sructwdingy element « the dashed tine in $iqc@) shows the oviginal seb for velewcnee,| and the solid tine shows the Twit beyond which any | fusther displacements of Ihe origin of B by 2 would cause the intersection of A ord A do be emply -rheackac, all poinds on and inside {his boundony, constitute the dilation of A by B. Fiy Cd) Shows a styuctuting clement designed 40 achieve move dilation vestically than hoxigentally, sand figtc) Shows the dilation achieved taith lbs element| | ; tly | \ I dq | | | : be | | | a lave | ala Ve 4 Np | f | | 7 ima | I lela r dig || of? : | | | ‘ 4d |) |_| = jo Bs bee. | | “| dg dag | | FIP * fa) sek a : , | (b) Squaxe stauctuaing, element . ~ \\ (0) dilation of A by B »shoum Shaded | (d) Elongated stauctusing, element (e7 Dilation of A si this element. the dotted sia im G) & (er 4s the boundary, of Set Ar Shown | only, foy vefesence. OM Geaombininge Dilali Combine Dilation ® Lyosiont- opening? and closing :- wee a pilalion expands the components of an frmage and exosion Shainks hem Tn this section we disc {wo alhex impoxtant morphological opevations : aperingy and closing Opening, genezallys smoothes the cota of an thin obsect , bycaks vaxow islhmuses , and ekiminales protrusions - closing, also tends fo smooth of contours bat , 08 opposed fo opering, f generally fuses MAvHOW byeaks and long, thin galls , elerminates srrall holes. and fills gaps in the contous the opening, of Set A by styuckuaing- © denoted AoB 15 defined as, lement 8, A 08 = (A@B)@B Thus, the openings A by B i the caosion of A by B , followed . by a dilation of the vesalt by B. gimilayly the closing of set A by glyuctusing element B ,denottd A+B 115 defined as, ° AeB = (A® BICB which says that Lhe closing, of A by 8 is simply, the dilation of A by B- followed by the crasion of the vesulé by B- ont “A Us/s346 the opening opewalion has a Simple qcormelaic mleyprclation » sappose thal the sévuctusing, element B as a (flat) 6 the boundary of a AoB 1 then in B that seach the farlhes¢ 1g volled avourd the inside we vicar “yollingy ball". established by the poinl indo the boundary of A 4% 8 of [his » boundany - this qeormebaic the openings opexation leads #0 4 which stotes that lhe opening, of slates of B that fit expyessed 05 filling, filling, Property of get theoretic forrnulalion, A by 8 15 ablained by faking. the union of all byan inl) & A. that 48 opening, Can be - pyocess Such that AoB = VS) Ite), cA $ whese Ufe} denotes the union of all the sets inside the byaces- closmg has a sim inteypzetation, lay qeometsic except that now we wll B on the outside of the e boundoay, + As shown below » opening, and closing, ane dual of each other » $0 having fo soll the tall on the outside is not unexpected - Gieomatyically , a Point w 16 an élemenk of A+B if and only if (B), NA +g for any translate of (B)z that contains w Figuae Q sllusteates the basic geometuical paoperties of closing . ONE, 6/53(oa | now OF Frgane aM: (a) 5tauctarey wUND. element Bou . boundary» of 4 rolling” alongs the ‘nme a (b) Shuctuaing, element ment (c) the heav: Ye line 5 ea ne is the outer boundary of the (cd) Complete opening, we did not shade A in (a) for Clasdty. Pv a Fig. @ @) Sbyactuwingy clement (8? “yolling "on dhe outer | boundayy of sel A (b) the heavy line 18 the outer boundaxy closing - (6) Complete for co clarity: of the | closing’. we did vot shade A in (a) unit by yals3ug Hil COR) Miss. Tans form’ the hit -@y) -miss dvansfoam 18 a general binoyy. Noxpholegical operation That can be used {0 look for Payticulay palleyns of foxegyound and background pirel, in an tmage «Il is actually, the basic apexation of binaay Mayphology* since almost all Lhe other binary Moxphologin| operaddys Can be desived formit: The sample Sbructuying, Masks &Sed are shown ‘n below figure. | the algosithrs for the hit -ox- Miss laansfoxm ‘is those of the othey Moxphalogical opezatoys Similax to 1 a3 follows. | ¥ Panslate the center of the stsuckuring, element to | ox all the pomts of the input smage. | % Compare the styuckweng, element with the tmage | pixels * Tf Ueve if a Complele match’ , then Ue piel underneath the sfauckuaingr element | ts sed to the forgyound coloux- this opevation is called a hid - nit) egg