Scribd
Upload
22 views

Dip unit 3

image reconstruction and restoration

Uploaded by

sriramkavi3
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
22 views

Dip unit 3

image reconstruction and restoration

Uploaded by

sriramkavi3
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
You are on page 1/ 23
: ONU-IL simage Récteratios: feconsteuctiont Je) of Image Beqracation [Restoration procese A Med | Image Restoration oe at 8 a precess of Recoveting on image that has been raided! sorne knowleahe of deqrocktion function 14 9 the achitive noise team now: Thus in yestoration, eqradation Tcl modelled § its invense process 1S applied to necoven the origiral Arnage A aaa Cuy) feasp | begodation ye mbneene | ot 5 may | Faretion(h) i. a roy aT Saas a Daqredation Restorotion Fee noge Restotantion % Degracetion Model the objective Ff rroge Restoration ts to obtain an estimate of Brginal image fouy) - tere, by Some knowleclge of HA youy, we fiocl appropriate Restoration Fillers: The principal Source of Neise In digital | images oxises dusung. Ie Tmage Acquistion, ea OS) Transmission Bethe peromance of ne isera is qjeced by a woxiety of -fhufronmental ¢, Mechonical = for instance, [n acquiving Images with a ced vee ate levels § Sensor temperatoxe are 3 Tmoges ane Compted esting -the transmission prececs ctie to interference in channel map =| 5 ai fon ) fT wo ae ails FS Noise qouy) The goal of Restoration filler is to obtain close replica Fouy) o Bagical Smage The fakennal Noise is probabilistic jn Nakwe § 4here ane several Noise medels used frequently in field of DIP. we have Several prcbibility density Funchons o| Noise. Noise Moclels +- > Gaussian Noise iat, Kayleigh Moise > Exlang(Gomma) Noise oa Exponential Notse —> Ontform Noise > Salt & pepper (amputlse) Noise- 4 aussian (Nirmal) Notset- 1 Used Frequently ra) practi Mathematical tiactablity regency domain. 2-eaussian Nvise cxtises 0 ce bemuse of its wn both spatial 9 the imoge due to Electronic circuit Noise Sensor Noise High tamperotone 8. palppr a Gaussian Rv, % iS ren by -(2-2) 2» P@ = papa ae | a a ‘Intensity J = mean value of = | g= Stondend deutstion J = c= vaniance Y # Apptor, 307% of Value tn Range (E-9), Ze] ‘ as,“ Se au cee [C2-20), (Z+205] * Kauletigh Nein t— ) t. Helpfal to choaaclexizing Noise phenomenon #0 | re 2. PDFS -@=0)* aE | Pee, ag ao | for 1 al e Pes eae The Mean vatue, The eno. of Density Function is [(b) - * Exponential Noise Finds application in lagen tmogin 2. This PDF is spect case Erlang POF ee bet. 2 PDF of Exponential Rv, 7 i P(z).= i ae fr 720 © % Uniform Noise lL Zt is quite useful fas basis Jor Numerous Rardom Neubert that used in Simulators - - to analyze System 2. PDF > 1 PE) -diece O22S6 Gal > * Tmpulse. Noise : 1. Found. tn situations Whete quick transients, take. place: cleuing Srraging pz) wis, ' g P, for z= nes P= is fot z=b | ° else 7 Pe Zz AP ba, intensity & will appear as light det in ‘image. ena a wtilt appecer like & bigek dot in image - ctence, this presence of Back § White | dots sesembles; 10»).Salt & pepprl granules. -tlence also calted Salt -and- pepper Noise. When etthn R B PR te zo, tk is Called Onipolon Niece. Rectoration iin the presence of Noise only Pa Spatial Elltering The na ts clegroclecl because of Noise only, The clegrodad medel can be fep- as Goupp = feu) + cy) Gewese Flav) + N(uwv) Spatal Filtering 7S a) Method o| Tikeing out oF reise “fom Images’ using spectfic choles. of Spatial Filters. Spatial. Filters ane used Rr mage. Restoration = Feroval e Noise. We cliscuss sevenat useful Filters Goren used im pip. pe Mean Filters Orckn - statistics Alters Aalaptive Filters Mean Fillers 1 Avithmelic Mean Alter 2 Geometric ie oe 3. Harmonic 4.Contra hasimonic Mean Filter }. Anitnmetic 2— + Computes the avenge value of Corrupted ‘rage ey fn the asea clfined by Suy: [sy set of Coordinates in Rectangular submage window of size mxn, centered at point (a4)- ‘The! eine of Restored image POu4) te fap: 5 gt) EOE iy * Smeothes Iocal variations in an image, and reise is Reduced as a Result of Blusering - 2 Gecmetotic <— +The value of Restored mage 1 fow= mut 0) es (50S, Bee "Each Reclored pixel i grer by product af pixels in Sulbimage inde » valsed to pocoer (mn Packives sreclhing than couthanct® man Fle “Bub teres io lese tess Trnoge details tn prose 3, Hosmonic 2- + Restored dmage a mo FOupe DES © Works lel! for salt Notse, fails fer pepe" * Works well hiith aul other Moises 4. Controhasimonic 2— * Restored “Image Qt) a Fionn sb the Behaviow of ochptive fitter changes deperctita on charactexfstics of ‘Image Inside the Alter Regio > The Medan Filter performs Relatively wlell on ‘Tmpulse noise as {eng as epatial density of impulse noise 1S pot ee > tle Will take a look at adaptive Median Filter > The Adoptive mection Filter can handle much more Spotiatly dense ‘impulse noise § also performs Some Smoothing fer non—impulse noise. ~ key to undeistancting the clgovithr -- * Remove ‘Impulse Noise + provicle Smoothing of other Noise * Reduce distortion @ ed Néfce Retention by Frequency aie ) | Let US Considen +he Following pastametens Emin = Minimurm intensity value tn Sx, Zmaxn = Maximum a f a Zmed = Median of si : j Zag = intensity values at Co-Adlinates (2,4) Sra = Monimrum cutotced size of S The Acbptive mectian pilkentng lieth waiks i tooo stages} denoted as stage A a G Stage B OS follotos y- “a =e 1g) a Al = Zmed ~2min A2 = Zmed —Zmax TP AI20 AND Ar<0, Pr StogeB Else ‘Trcrease ’ the windxo size aF wirckeo size Else oukput Cae tes stoge A singe B= BLE Zxy- Zein Bo = Zay ere | # Bize AND. 62.20, cutpub Zay Else catpat reel ®@ ee Noise eduction By Frequency lomein Filtening pericalic Noise in images ane oppaned ac concentrated bursts of energy ih Foundes ‘tansjorm at locations corres pordin to the frequencies of- petiodic ‘interference. This’ Can be removeol by using selective Filten. © Band Reject Filters | ) ®eord pass Filteis — © Notch Filters © and Reject tiller. ! BP Atcencte (Reroves a bard bf Frequencies afove GIR the Bigin. of ET Buigi f i » Wdeat :- ibe DOM) = Bor > i A w Ooty pe Dv) =e" 2 Hv) = w, Wes Dwv) 2 Bt > req Eond DU) ctictance from Bligin of the centered f — W width a bond Bo > Radiat center / cutoff bey Buttoworth t- rr eburnie J H(OV) = 7 ee 7 [ 0 (ov)-D> Eaussian 2— mt [pte oO] PO Enemies filter 20 © Band pass Filter : E performs the opposite operation of Bard reject Filler A Band poss Filter & obtained prom the bord reject File. as Hpp(V) = t— Hpp (Uv) © notch Eilter 2 (noth Filter (Reject 6 passJ) passes Frequencies i predefined Neighbours Tt is constowcted as product of high pess Fitas whose centers hove been | towanslated to Centers of Notches. The Genesiat Form i¢ is Hyp (Ur) = TT Helo) cu, v) ca hihere Hy (u,v) 8 Hylorv) ane high pass Filters those centers ane at (Her) 4 (ete ee) ME ; Respectively ‘ A Notch pres Filton obtained -fém Notch Reject CMR) Hyp w= '— Hye (ov): radotions 6 Linear, Pasitions ~ Dnvastiant the (he Tp -clp Relation Relationship —bejsie €storation stage is expressed as qmy = 4 Loo p+ cow ie. in absence of adlikive Moise i =--@) Olet us assume gcxiy)=O/ gens HLCup) — © FO iF scalon votes of ag b, 4 18 Linear if 4 fateep + batoy]= oH [Row+ pufhoyy] 4 3 a=b=15 TAT ps hoon) * Afiayle nite) Which % called the property Te Thus a Linean operator possess both tte aw ‘i adttivity & property of homogeneity | RUA awe H[af coy] = an [hoop] wikich is called property | ok! @ An qprior having tip ~ op IP qouyp= HEFOuy) iS said to be position invanienE if ‘ (space) HE@-w, y-p= Yeu YP © Foy) can be expresarcl in terms of Continuous } tivity. With 401.¥xJ ‘impluce —-feinction ( ooyy= | [-ReuB) Stam, yo p) eet <0 -d tad, lee Co -Ampe aes qoop = HFouyi] , sul] Sree) SG yp) dep) © x 1 Linean eperator % we extel the additivit propety to ntegrals , then r Gey) =H[Fonyi) = J fafre, p) SO d,y Bicep 00 iar, 0) Mere FB) tS indeperdent of % y and using homogenity prety: JOup = H[Fougi] = i f FB) H[Stx-03,4-B)] dudp 00-0 h(x, 4) BY= HE SG-w, y-B “this is called -Imputse weSponse of H TE 10H4)= 0, then h(neiy B) is Suspense of H “to an impulse Steength Lak cobldinates (4) lu YB) = H [SOR 4B] H[BG«y-BY] = h(a YB) 0 [gen = ‘i { eB) h@-aiy-P) dt lB Ee a 2 ee a= § J FEN CHey-p)dedp + 2Cug) | BP ee hoay= H1SG.4)) © Ectimating -the leqrachtion Fanction Thee are Three principal woys to estimate the degradation function for use in image “restoration Given os foltaws — Estimation y Lmoge observation eee by " expeumentation = “Mathematical Mecleting The prowess of vestovi ean Tenage. by usin: or deqrad— ation Function that has been estimated’ in some way sorrelimes 1S called Blind deconvolution. # Orgtimation by mage _Cbsowation® Jn this pxess , the degradation function iS : estimated by obsewing: the . Select the Sub image Whose signal content ts Stowng. Leb the * obsewed
    | Foy = ftww= Eloy NON How), | Bom the above expression, we can obseve tok Ff te crow thek degrade Function pte canct Ga x TecOver the tn — degrodect imoge exactly because the Noe N(O,v) 7S Nlot known. re the | degrade ferction has zero & eur Small values, | then the rato Nov) / H(o,v) cou Id easily dominate the B (uy). to avid thts dicod- evontage 1 We limit the filter brequency Values Nan the Bigin becouse Hoy) values ane Moximem ab Le buigin D ‘Minimum mean Squane error (wiener)_Eillering dn tne Hltediog process it incorpoyates both the deqrode Furchion ¢ statistical characteristics of Notse- the ‘images g Noise in this Methed ane coreicered 0 Random variables . she objective is to find an estimate f of the uncorrupted image such that the mean square error bli them ie minimized. This ewoy is meacuned by 4 ae), suhoe EL 5 6 expected value of argument - > Tt 1 assumed int Noise § droge axe unconel, ed. > Gitur Nice & Imoge has cero mean~ => The Intensity levels in estinate ane a linear Fanction oj levels in degraded image. 4 Based on these condition, the Minimum of ens? punction th —frecqprency domain is given as POV) = | Ato) {: HOW) Jr(u wl Sr (ow) [Ss Heo) Seo) slow) |HlovyI? + S_Corv) Jee) __ Gee | eee ions IS \ H(ovl” y —~— | & (ov) HOV) = Degraded function Huy) = complex Conjugate of Hlow) Riel’ = Hlov) Hew) Sylvy Vv) Sp(orv) ao) INtoNlh= poet Spectem Of er ewe « s Undegraded Trans $800 of deeyroded 4mage thé Result E(ov) 7 knocon as usienen Filter. The toms inside the brackeE TS Commonly referred G8 miinimun mean square evi) fitter 8) least equane ero filta. EE doesnot have Same problem aS jnvese Riter hlith ZEIOS in degraded Function, unless - the’ entive olenommota tS toro aP moe is tow, then the Nose pocsen Spectwn, Vanishes § wien Filten reduces to thee Fite. PA No. & usefl measures ane available tov evaluation “ome artong” than 8 signal td Nowe Ratto to bel The siqrol nO ee fo in freq domain a Eco lls ae Cc O cme we eel . pis crv Gee Vz8 The SNR in Seabial domain Daa Coase SNR = SE, FE imp Fey)” The mean sq. exer 1S Obtained by using He op Mot MSE = ae Fep- foo)” H=OYFO The Modified exp to estimate z by using Minimum MSE filtewng es 1 LHfow)) > a 20 | on Taps kh” @ Constratne d_ least Sourore _Filteuing: “the bin with ntiener Filter 7s that if is necessary to Know the power spectrum of Noise § dmage. The degradec! Ymage in spatiat domain is Jey = LOOY)* houyy + yOuy) Zin vector- maton porn qe Hf $1 £ where gin uectors of climensions MNxl, H matrix of dimension MNX MAI iS VOW lange & is highly sensitive to Noise - Optimality ob restoration baged on & measune of Smoothness < Usi laplecian: operator - The vestor- -ation must be constained by iametens to pind minimum criterion punctions's defined ac ia Meal gict c= DS Pv Houyy) Cr Second ad ae dsivative ) From degradation ap eeu she criterion Function 15 sbbject to constraint wg-Hfh= [bee whe fi an estirate 9) undeqaded Image The eq domain soln to wis optimization prblem 6 given by Ht (Ov) A . SaNRETSA ee Be Mie [ Ine! + yeowl hithere, -Y is porrometer thot mast be adjusted 0 thot constoint is satishied- -] Giuv) | | © Geometric Meo Mean Filter. PLOW oN Pm ct Ecneson eee POU (=I aa at is possible to adjust the poo jamentn “Y Hter actively until occeptable aeailts ane achteved, 31 use mean § voniorce of Weise to ibeatively ° adpsE Y (40 satisfy critexian function constratnt ) the vantance ¢% mean iS i I Not > a Ne ra) ea ae [feet er 1a (uy) ae nl ng au Hence noice is ll ne MN [enema | +TE © posite -to generalize slightly the wieney-Filtey + the Bee ont is in the form of the so-called geometouc mean Filten. ol I> @ fow)= ey meg ’ lH vp Intow)le B Sn (ev) oe Sp(OrV) 4 WhEx 2B ane beg posttive g meal constants > The geometxic mean filter consists of two expressions in brackets naised to powers « § la mespectivel > when cet, this Filter vedices to the inverse fille. Puen x0, “becomes the sd-Calted Povemetfic tienen Filter, which vedkices to Stondand = ytiem> = Eten Wihen pei Suthe X= +, the Filter veduce becomes a produc of two quantites saised to same pocwen, which © the definition of geometouc mean, thus giving the filter its nome eeometiic mean Filter. > ltith pel a3 & decreases below Y> , the Filter perpaimerce. will tencl move toward the inverse diel Fee rerence frbove Yo, the Filter uit behoye mie like culenen Filter = when a=Y2 % Bel the Filken alsots Commonly refered tb as Spectoum equalization filken = Thus, eqn shoun 1s quite useful bhen jimpleme— nting vegoration filters because it veally vepresents % fomilty o, biltere Combined ‘into a Single expression] uthere ~ | B being +ve veal Constants, Based on Nalues of x5 B , geometoiic mean filter pedis the difkeount actors 1 = invese Filler fs i ike =O => poriametouc wliener Fil é (stanclastel « we athen pel) d= YY. S actuat Ceometoiic mean. b=Y> § pl > Spectoum equalization Filter

You might also like