66
© 1992–2008 R. C. Gonzalez & R. E. Woods Chapter 5: Image Restoration and Reconstruction Degradation process: Objective of restoration: To reconstruct original image as accurately as possible Degradation and restoration model: Degradation function H: normally linear and shift invariant system

Degradation process: Objective of restoration: To reconstruct …users.encs.concordia.ca/~weiping/ELEC6641-W14/Chapter 5.pdf · 2014. 3. 19. · Chapter 5: Image Restoration and Reconstruction

  • Upload
    others

  • View
    1

  • Download
    0

Embed Size (px)

Citation preview

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Degradation process:

    Objective of restoration: To reconstruct original image as accurately as possible

    Degradation and restoration model:

    Degradation function H: normally linear and shift invariant system

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    1. The restoration in general is a complex process. It depends on the distortion model H and the nature of the noise.

    2. If there is no distortion, then restoration becomes a denoising problem.

    3. Usually, it is assumed that the noise is independent of the image, and noises at different pixels are also independent.

    4. The signal to noise ratio is defined as

    5. Peak signal to noise ratio is defined as

    2

    10 10210log =20log dBf f

    n n

    SNRσ σσ σ

    =

    10 1025520log =20log dBpeak pixel valueSNR

    η ησ σ

    =

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    (a) A digital image with standard deviation 55 , (b) the same image added with a noise of standard deviation 3 (making SNR to 25.26 dB), (c) the same image with SNR around 6 dB.

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Different Noises

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Gaussian noise:

    Rayleigh noise:

    Gamma noise:

    −−= 22

    2 2)(exp

    21)(

    σµ

    πσxxp

    4/ba πµ +=4

    )4(2 πσ −= b

    an=µ 22 an=σ

    a > 0, n is a positive integer.

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Test Pattern

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Noise Corrupted Test Pattern

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Noise Corrupted Test Pattern (Cont’d)

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Sine Noise Corrupted Image

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Noise Corrupted Strip Image

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Noise reduction – spatial filtering

    1 Mean filters

    Arithmetic mean filter

    ),(),(),( yxyxfyxg η+= ),(),(),( vuNvuFvuG +=

    ∑∈

    =xySts

    tsgmn

    yxf),(

    ),(1),(ˆ

    Geometric mean filter

    Harmonic mean filter

    mn

    Sts sy

    tsgyxf

    1

    ),(

    ),(),(ˆ

    = ∏

    ∑∈

    =

    xySts tsg

    mnyxf

    ),( ),(1),(

    ˆ

    Contraharmonic mean filter

    +

    =

    xy

    xy

    Sts

    QSts

    Q

    tsg

    tsgyxf

    ),(

    ),(

    1

    ),(

    ),(),(ˆ

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    2 Order-statistic filters

    Median filter

    Max and Min filters

    Midpoint filter

    Alpha-trimmed filter

    { }),(),(ˆ),(

    tsgmedianyxfxySts ∈

    =

    { }),(max),(ˆ),(

    tsgyxfxySts ∈

    =

    { }),(min),(ˆ),(

    tsgyxfxySts ∈

    =

    { } { }

    +=

    ∈∈),(min),(max

    21),(ˆ

    ),(),(tsgtsgyxf

    xyxy StsSts

    ∑∈−

    =xySts

    r tsgdmnyxf

    ),(),(1),(ˆ

    Remove dark pixels

    Remove white pixels

    Remove salt-pepper noise

    Remove outliers by excluding d/2 brightest and darkest pixels

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    2

    2

    2

    2

    2 2

    2 2 2 2

    ( , ) ( , ) ( ( , ) )

    Local mean

    Noise variance

    Local variance

    If (less noise), then

    If (more noise), then and i.e. when there is lo

    NL

    L

    L

    N

    L

    N L

    N L N L L

    f x y g x y g x y m

    m

    f g

    f m

    σσ

    σ

    σ

    σ σ

    σ σ σ σ

    = − −

    = =

    =

    =

    ≈ ≈t of noise, we just average it over pixels.

    ∑∈ ),(),(

    ),( 1yxSts

    tsgmn

    3 Adaptive filters

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Local noise reduction filter

    Pseudo code for adaptive median filter:

    [ ]LL

    myxgyxgyxf −−= ),(),(),(ˆ 22

    σση

    Stage A: A1=zmed-zmin A2=zmed-zmax If A1>0 and A20 and B2

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Arithmetic and Geometric Filtering

    Guassian noise added

    3X3 Averaging Geometric mean filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Contraharmonic Filtering

    (a) Pepper noise

    (b) Salt noise

    Q=1.5 Q=-1.5

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Q=1.5 Q=-1.5

    Wrong Selection of Parameter

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    (a) Pepper & salt noise

    (b) 3X3 Median

    Order Statistic Filtering

    2nd Median Filtering 3rd Median Filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Results of Max & Min Filtering

    Filtering of pepper noise corrupted image

    3X3 Max filtering 3X3 Min filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Mean Filtering & Order Statistic Filtering

    Uniform noise

    5X5 Arithmetic

    Median filtering

    Uniform + salt & pepper noise

    Geometric mean

    Trimmed mean filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Filtering of Gaussian Noise with m=0

    Gaussian noise added Arithmetic mean filtering

    Geometric mean filtering Adaptive median filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Salt & pepper corrupted Median filtering

    Adaptive median filtering

    Median and Adaptive Median Filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Frequency Selective Filters

    Ideal bandstop filter

    Butterworth bandstop filter

    Gaussian bandstop filter

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Removal of Sine Noise

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Sine Noise Image

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Notch Filters

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Notch Filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Filtering process:

    1

    ( , ) { ( , )},( , ) {[1 ( , )] ( , )}NR

    G u v F g x yx y F H u v G u vη −

    =

    = −

    ),(),(),(),(ˆ yxyxwyxgyxf η−=

    [ ]∑∑−=−=

    −++++

    =b

    bt

    a

    as

    yxftysxfba

    yx2

    2 ),(ˆ),(ˆ)12)(12(

    1),(σ

    To determine w(x,y), consider

    ∑∑−=−=

    ++++

    =b

    bt

    a

    astysxf

    bayxf ),(ˆ

    )12)(12(1),(ˆ

    Where is average of in the neighborhood.

    ),(ˆ yxf),(ˆ yxf

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    2

    2

    ]),(),(),([

    )],(),(),([

    )12)(12(1),( ∑∑

    −=−=

    −++++−++

    ++=

    b

    bt

    a

    asyxyxwyxg

    tysxtysxwtysxg

    bayx

    η

    ησ

    ),(),(thatAssume yxwtysxw =++

    ),(),(),(),( yxyxwyxyxw ηη =2

    2

    )],(),(),([

    )],(),(),([

    )12)(12(1),( ∑∑

    −=−=

    −++−++

    ++=

    b

    bt

    a

    asyxyxwyxg

    tysxyxwtysxg

    bayx

    η

    ησ

    By setting 0),(),(2=

    ∂∂

    yxwyxσ

    ),(),(

    ),(),(),(),(),( 22 yxyx

    yxyxgyxyxgyxwηη

    ηη

    −=

    we obtain

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Modeled by PSF

    ),(ˆ),(),(

    vuFvuGvuH

    s

    ss = A

    vuGvuH ),(),( =

    Estimation degradation function

    6/522 )(),( vukevuH +−=

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Air Turbulence Model

    K=0.0025

    K=0.001 K=0.00025

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Motion Blur Model

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    ∫ −−=T

    dttyytxxfyxg0

    00 )](),([),(

    Motion blurring model

    [ ]

    ),(),(),(

    )](),([),(

    0

    )]()([2

    )(200

    0

    00 vuFvuHdtevuF

    dtdxdyetyytxxfvuG

    Ttvytuxj

    vyuxjT

    ==

    −−=

    ∫ ∫∫

    +−

    +−∞

    ∞−

    ∞−

    π

    π

    dtevuHT

    tvytuxj∫ +−=0

    )]()([2 00),( π

    Further assuming linear motion in both x and y directions: TbttyTattx /)(/)( 00 ==

    )()](sin[)(

    ),( vbuajevbuavbua

    TvuH +−++

    = πππ

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Inverse filtering

    ),(),(

    ),(),(),(

    ),(ˆvuHvuN

    vuFvuHvuG

    vuF +==

    For example

    6/522 ])2/()2/[(),( NvMukevuH −+−−=

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Inverse Filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Wiener (MMSE) filtering { }22 )ˆ( ffEe −=

    ),(),(/),(),(

    ),(),(

    1

    ),(),(/),(),(

    ),(),(),(),(),(

    ),(),(),(ˆ

    2

    2

    2

    *

    2

    *

    vuGvuSvuSvuH

    vuHvuH

    vuGvuSvuSvuH

    vuHvuGvuSvuHvuS

    vuSvuHvuF

    f

    ff

    f

    +=

    +=

    +=

    η

    ηη

    ∑∑

    ∑∑

    ∑∑

    ∑∑−

    =

    =

    =

    =−

    =

    =

    =

    =

    −≈= 1

    0

    1

    0

    2

    1

    0

    1

    0

    2

    1

    0

    1

    0

    2

    1

    0

    1

    0

    2

    )],(ˆ),([

    ),(ˆ

    ),(

    ),(

    M

    x

    N

    y

    M

    u

    N

    vM

    u

    N

    v

    M

    u

    N

    v

    yxfyxf

    yxf

    vuN

    vuFSNR

    ),(),(

    ),(),(

    1),(ˆ 2

    2

    vuGKvuH

    vuHvuH

    vuF

    +≈ ∑∑

    =

    =

    −=1

    0

    1

    0

    2)],(ˆ),([1M

    x

    N

    yyxfyxf

    MNMSE

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Inverse and Wiener Filtering

    Full inverse filtering Radially limited filtering Wiener filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Motion Blur and Additive Noise

    Corrupted image

    Inverse and Wiener filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Constrained LS Filtering

    ηHfg +=Degraded image in matrix form:

    Minimize [ ]∑∑−

    =

    =

    ∇=1

    0

    1

    0

    22 ),(M

    x

    N

    yyxfC

    Subject to 22ˆ ηfH-g =

    ),(),(),(

    ),(),(ˆ 22*

    vuGvuPvuH

    vuHvuF

    +=

    γ

    where P(u,v) is FT of

    −−−

    −=

    010141

    010),( yxp

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Constrained LS Filtering

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Constrained LS Filtering with iteration of

    Using correct noise parameters

    Using wrong noise parameters

    γ

    γ

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Projection, back projection, and superimpose

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Reconstruction using back projections

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Reconstruction using different projections

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Reconstruction using back projections

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    ∫ ∫∞

    ∞−

    ∞−

    −+= dxdyyxyxfg jkkkj )sincos(),(),( ρθθδθρ

    ∫ ∫∞

    ∞−

    ∞−

    −+= dxdyyxyxfg )sincos(),(),( ρθθδθρ

    ∑∑−

    =

    =

    −+==1

    0

    1

    0)sincos(),(),(),(

    M

    x

    N

    ykj yxyxfgg ρθθδθρθρ

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    ≤+

    =otherwise 0

    x),(

    222 ryAyxf

    dyyfdxdyxyxfg ∫∫ ∫∞

    ∞−

    ∞−

    ∞−

    =−= ),( )(),(),( ρρδθρ

    dyAdyyfgr

    r

    r

    r ∫∫−

    −−

    −−==

    22

    22

    22

    22),(),(

    ρ

    ρ

    ρ

    ρρθρ

    ≤−

    ==otherwise 0

    r 2)(),(22 ρρρθρ rAgg

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    ),ysing(xcos),(),( kkk θθθθρθ +== kgyxf k),sincos(),( θθθθ yxgyxf +=

    ∫=π

    θ θ0

    ),(),( dyxfyxf

    ∑=

    θθ

    0),(),( yxfyxf

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    ∫∞

    ∞−

    −= ρθρθω πωρdegG j2),(),(

    dxdyeyxf

    dxdydeyxyxfG

    yxj

    j

    )sincos(2

    2

    ),(

    )sincos(),(),(

    θθπω

    πωρ ρρθθδθω

    +−∞

    ∞−

    ∞−

    ∞−

    −∞−

    ∫∫

    ∫ ∫ ∫

    =

    −+=

    [ ] cos ; sin( , ) ( , )( cos sin )

    u vG F u v

    Fω θ ω θ

    ω θ

    ω θ ω θ= =

    =

    = +

    FT of projection:

    Reconstruction of back-projected image using sinograms

    ~ Fourier slice

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Reconstruction using parallel-beam filtered backprojection

    ∫ ∫∞

    −∞−

    += dudvevuFyxf vyuxj )(2),(),( π

    ∫ ∫∞

    +=π

    θθπω θωωθω2

    0 0

    )sincos(2),(),( ddeGyxf yxj

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    ∫ ∫∞

    ∞−

    +=π

    θθπω θωθωω0

    )sincos(2),(),( ddeGyxf yxj

    θωθωωθθρ

    ππωρ ddeGyxf

    yx

    j

    sincos0

    2),(),(+=

    ∞−∫ ∫

    =

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    To get complete and back-projected image: 1) Compute 1-D FT of each projection 2) Multiply each FT by the filter function which has

    already multiplied by a suitable window 3) Obtain the inverse 1-D FT of each resulting filtered

    transform 4) Integrate (sum) all the 1-D inverse transforms from

    step 3).

    ||ω

    ≤≤

    −−+

    = otherwise 0

    1)-(M0 1

    2cos)1()(

    ωπωω M

    cch

    Frequency-domain windowing function:

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Reconstruction using fan-beam filtered backprojection

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    αβθ += αρ sinD=

    θρρθθθρπ

    ddyxsgyxfT

    T∫ ∫ −+=−

    2

    0

    )sincos(),(21),(

    )cos(sinsincoscossincos

    ϕθθϕθϕθθ

    −=+=+

    rrryx

    [ ] θρραθθρπ

    ddrsgyxfT

    T∫ ∫ −−=−

    2

    0

    )cos(),(21),(

    [ ] βαααϕαβ

    βααϕαπ

    α

    ddDDrs

    DgrfDT

    DT

    cossin)cos(

    ),sin(21),(

    2 )/(sin

    )/(sin

    1

    1

    −−+

    += ∫ ∫−

    − −

    Let be the IFT of ( )s ρ | |ω

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    [ ] βαααϕαββαϕαπ

    α

    α

    α

    ddDDrsprfm

    m

    cossin)cos(),(21),(

    2

    −−+= ∫ ∫−

    − −

    )sin(sin)cos( ' αααϕαβ −=−−+ RDr

    [ ] βααααβαϕαπ

    α

    α

    α

    ddDRsprfm

    m

    cos)'sin(),(21),(

    2

    −= ∫ ∫−

    − −

    )()sin

    ()sin( 2 αα

    αα sR

    Rs = βααβαϕα

    α

    π

    dhqR

    rfm

    m

    −= ∫∫

    )'(),(1),(2

    02

    )(sin2

    1)(2

    αα

    αα sh

    = αβαβα cos),(),( Dpq =

    ),sin(),(),( βααθρβα +== Dggp

    γαβ =∆=∆ [ ]γγγγ )(,sin),( nmnDgmnp += If

    The n-th ray in the m-th radial projection is equal to the n-th ray in the (m+n)-th parallel projection!

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

  • © 1992–2008 R. C. Gonzalez & R. E. Woods

    Chapter 5: Image Restoration and Reconstruction

    Slide Number 1Slide Number 2Slide Number 3Slide Number 4Slide Number 5Slide Number 6Slide Number 7Slide Number 8Slide Number 9Slide Number 10Slide Number 11Slide Number 12Slide Number 13Slide Number 14Slide Number 15Slide Number 16Slide Number 17Slide Number 18Slide Number 19Slide Number 20Slide Number 21Slide Number 22Slide Number 23Slide Number 24Slide Number 25Slide Number 26Slide Number 27Slide Number 28Slide Number 29Slide Number 30Slide Number 31Slide Number 32Slide Number 33Slide Number 34Slide Number 35Slide Number 36Slide Number 37Slide Number 38Slide Number 39Slide Number 40Slide Number 41Slide Number 42Slide Number 43Slide Number 44Slide Number 45Slide Number 46Slide Number 47Slide Number 48Slide Number 49Slide Number 50Slide Number 51Slide Number 52Slide Number 53Slide Number 54Slide Number 55Slide Number 56Slide Number 57Slide Number 58Slide Number 59Slide Number 60Slide Number 61Slide Number 62Slide Number 63Slide Number 64Slide Number 65Slide Number 66