Low Pass and High Pass Filters

  • Upload
    src0108

  • View
    212

  • Download
    0

Embed Size (px)

Citation preview

  • 7/29/2019 Low Pass and High Pass Filters

    1/6

    Dr. D. J. Jackson Lectur e 10-1Electrical & Computer Engineering

    Computer Vision &

    Digital Image Processing

    Frequency Domain Filters

    Dr. D. J. Jackson Lecture 10-2Electrical & Computer Engineering

    Butterworth lowpass filter

    The transfer function of aButterworth lowpass filter (BLPF)of order n with cutoff frequency

    D0 is given by

    where D(u,v)=[u2+v2]1/2

    For this smooth transition filter, acutoff frequency locus is chosensuch that D(u,v) is a certainpercentage of its maximum

    Designed such that at D(u,v)=D0H(u,v)=0.50 (50% of its maximumvalue)

    [ ] nDvuDvuH

    2

    0/),(1

    1),(

    +=

    H(u,v)

    D(u,v)/D0

    Dr. D. J. Jackson Lectur e 10-3Electrical & Computer Engineering

    Butterworth lowpass filter (continued)

    Another transfer function ofa Butterworth lowpass filter(BLPF) of order n withcutoff frequency D0 is givenby

    Designed such that atD(u,v)=D0

    nDvuDvuH

    20]/),(][12[1

    1),(

    +=

    2

    1),( =vuH

    H(u,v)

    D(u,v)/D0

    Dr. D. J. Jackson Lecture 10-4Electrical & Computer Engineering

    MATLAB Butterworth lowpass filter

    f unct i on [ g]=bl pf ( f , or der, cutof f );

    % Usage [ g]=bl pf (f , order, cutof f );

    F=f f t2(f ) ;

    F=f f tshi f t( F);

    [ umax vmax] =si ze( F) ;

    f or u=1: umax

    f or v=1: vmax

    H( u, v) =1/ ( 1+( sqrt ( 2)- 1)*( sqrt ( ( ( umax/ 2-( u- 1)) . 2+( vmax/ 2-( v-1)) . 2)) / cutof f ). ( 2*order) ) ;

    end;

    end;

    G=H. *F;

    G=i f f t 2(G) ;

    g=sqr t ( r eal ( G) . 2+i mag(G). 2);

    Dr. D. J. Jackson Lectur e 10-5Electrical & Computer Engineering

    Example MATLAB output

    Cutoff frequency=10 Cutoff frequency=100

    Dr. D. J. Jackson Lecture 10-6Electrical & Computer Engineering

    Butterworth lowpass filter function

  • 7/29/2019 Low Pass and High Pass Filters

    2/6

    Dr. D. J. Jackson Lectur e 10-7Electrical & Computer Engineering

    Ringing in BLPF

    Dr. D. J. Jackson Lecture 10-8Electrical & Computer Engineering

    Gaussian lowpass filter

    The transfer function of a Gaussian lowpass filter (GLPF) isgiven by

    Here, is a measure of spread about the center Let =D0, then

    where D0 is the cutoff frequency

    22

    2/),(),(vuDevuH =

    20

    2 2/),(),( DvuDevuH =

    Dr. D. J. Jackson Lectur e 10-9Electrical & Computer Engineering

    Gaussian lowpass filter function

    Dr. D. J. Jackson Lecture 10-10Electrical & Computer Engineering

    Ideal highpass filter (IHPF)

    A transfer function for a 2-D ideal highpass filter (IHPF) isgiven as

    where D0 is a stated nonnegative quantity (the cutofffrequency) and D(u,v) is the distance from the point (u,v) tothe center of the frequency plane

    >

    =

    0

    0

    Dv)D(u,if1

    Dv)D(u,if0),( vuH

    22),( vuvuD +=

    v

    u

    H(u,v)

    Dr. D. J. Jackson Lecture 10-11Electrical & Computer Engineering

    Butterworth highpass filter

    The transfer function of aButterworth highpass filter(BHPF) of order n with cutofffrequency D0 is given by

    where D(u,v)=[u2+v2]1/2

    For this smooth transition filter, acutoff frequency locus is chosensuch that D(u,v) is a certainpercentage of its maximum

    Designed such that at D(u,v)=D0H(u,v)=0.50 (50% of its maximumvalue)

    [ ] nvuDDvuH

    2

    0 ),(/1

    1),(

    +=

    Dr. D. J. Jackson Lecture 10-12Electrical & Computer Engineering

    Butterworth highpass filter (continued)

    Another transfer function ofa Butterworth highpassfilter (BHPF) of order n withcutoff frequency D0 is givenby

    Designed such that atD(u,v)=D0

    nvuDDvuH

    20 )],(/][12[1

    1),(

    +=

    2

    1),( =vuH

  • 7/29/2019 Low Pass and High Pass Filters

    3/6

    Dr. D. J. Jackson Lecture 10-13Electrical & Computer Engineering

    Gaussian highpass filter

    The transfer function of a Gaussian highpass filter (GHPF) isgiven by

    Here, is as in the Gaussian lowpass case Let =D0, then

    where D0 is the cutoff frequency

    22

    2/),(1),(vuDevuH =

    20

    2 2/),(1),( DvuDevuH =

    Dr. D. J. Jackson Lecture 10-14Electrical & Computer Engineering

    Highpass filter functions

    Dr. D. J. Jackson Lecture 10-15Electrical & Computer Engineering

    Unsharp masking, highboost filtering, and high-

    frequency-emphasis filtering

    Let

    where

    Where HLP(u,v) is a lowpass filter andF(u,v) is theFourier transform of f(x,y)

    Then

    If k=1 this is an unsharp mask

    If k>1 this is a highboost filter

    ),(),(),( yxfyxfyxg LPmask =

    )],(),([),( 1 vuFvuHyxf LPLP=

    ),(*),(),( yxgkyxfyxg mask+=

    Dr. D. J. Jackson Lecture 10-16Electrical & Computer Engineering

    Unsharp masking, highboost filtering, and high-

    frequency-emphasis filtering

    In frequency domain only terms

    This is a high-frequency-emphasisfilter

    A more general form is

    Here k10 controls an offset from the origin and k20 controlscontributions of high frequencies

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

    filterhighpassaofin termsor

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

    1

    1

    vuFvuHkyxg

    vuFvuHkyxg

    HP

    LP

    +=

    +=

    )},()],(*{[),( 211 vuFvuHkkyxg HP+=

    Dr. D. J. Jackson Lecture 10-17Electrical & Computer Engineering

    Example use of a high-frequency-emphasis filter

    Dr. D. J. Jackson Lecture 10-18Electrical & Computer Engineering

    Homomorphic filtering

    Recall f(x,y) can be expressed as f(x,y)=i(x,y)r(x,y)

    We cannot use this directly to operate on the frequencycomponents of i(x,y) and r(x,y) because

    But if we define

    then

    )},({)},({)},({ yxryxiyxf

    )],(ln[)],(ln[

    )],(ln[),(

    yxryxi

    yxfyxz

    +=

    =

    )]},({ln[)]},({ln[

    )]},({ln[)},({

    yxryxi

    yxfyxz

    +=

    =

  • 7/29/2019 Low Pass and High Pass Filters

    4/6

    Dr. D. J. Jackson Lecture 10-19Electrical & Computer Engineering

    Homomorphic filtering (continued)

    Then,

    where I(u,v) and R(u,v) are the Fourier transforms of ln[i(x,y)]and ln[r(x,y)] respectively

    Z(u,v) can be processed by a filter function

    where S(u,v) is the Fourier transform of the result

    In the spatial domain,

    ),(),(),( vuRvuIvuZ +=

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

    ),(),(),(

    vuRvuHvuIvuH

    vuZvuHvuS

    +=

    =

    )},(),({)},(),({

    )},({),(11

    1

    vuRvuHvuIvuH

    vuSyxs

    +=

    =

    Dr. D. J. Jackson Lecture 10-20Electrical & Computer Engineering

    Homomorphic filtering (continued)

    If,

    then

    Taking the exponential yields the final result

    )},(),({),('

    and)},(),({),('1

    1

    vuRvuHyxr

    vuIvuHyxi

    =

    =

    ),('),('),( yxryxiyxs +=

    ),(),(

    )],('exp[*)],('exp[

    )],(exp[),(

    00 yxryxi

    yxryxi

    yxsyxg

    =

    =

    =

    Dr. D. J. Jackson Lecture 10-21Electrical & Computer Engineering

    Homomorphic filtering (continued)

    The process can be viewed graphically as above

    The illumination of an image is generallycharacterized byslow spatial variations (associated with the low frequenciesof the Fourier transform of the logarithm)

    The reflectance of an image tends to vary abruptly,especially at the junctions of dissimilar objects (associatedwith the high frequencies of the Fourier transform of thelogarithm)

    ln

    f(x,y)

    FFT H(u,v) FFT-1 exp

    g(x,y)

    Dr. D. J. Jackson Lecture 10-22Electrical & Computer Engineering

    Homomorphic filtering (continued)

    The filter function H(u,v) should/will affect the low-and high-frequencycomponents in different ways and can be approximated by

    If the filter function chosen is such thatL1 then the lowfrequencies tend to be decreased and the high frequencies are amplified

    functiontheofslopethecontrolswhere

    ]1)[(),( ]/),([20

    2

    c

    evuH LDvuDc

    LH +=

    Dr. D. J. Jackson Lecture 10-23Electrical & Computer Engineering

    Homomorphic filtering (example)

    Dr. D. J. Jackson Lecture 10-24Electrical & Computer Engineering

    Homomorphic filtering (example)

  • 7/29/2019 Low Pass and High Pass Filters

    5/6

    Dr. D. J. Jackson Lecture 10-25Electrical & Computer Engineering

    Selective fil tering

    Previously discussed filters operate over the entirefrequency rectangle (i.e. complete representation in

    the frequency domain) Occasionally it is useful to operate on specific

    frequency bands or small regions of the frequencyrectangle

    Bandrejector Bandpass filters operate on specificfrequency bands

    Notch filters operate on small regions of thefrequency rectangle

    Dr. D. J. Jackson Lecture 10-26Electrical & Computer Engineering

    Bandreject filters

    Bandreject filters can, in general, be easilyconstructed using the same concepts as described

    for other filters Assume the following:

    D(u,v) is the distance from the center of the frequencyrectangle

    D0 is the radial center of the band of interest

    Wis the width of the band of interest

    Dr. D. J. Jackson Lecture 10-27Electrical & Computer Engineering

    Bandreject filters

    Ideal bandreject filter

    Butterworth bandreject filter

    Gaussian bandreject filter

    +=

    otherwise122

    if0),( 00

    WDD

    WD

    vuH

    n

    DD

    DWvuH

    2

    20

    21

    1),(

    +

    =

    220

    2

    1),(

    = DWDD

    evuH

    Dr. D. J. Jackson Lecture 10-28Electrical & Computer Engineering

    Bandpass filters

    Bandpass filters can be derived from any of thebandrejectexpressions as

    HBP(u,v)=1- HBR(u,v)

    HBR(u,v) is the corresponding bandrejectfilter

    This formulation is exactly as in thehighpass/lowpass case

    Dr. D. J. Jackson Lecture 10-29Electrical & Computer Engineering

    Gaussian filter example

    Dr. D. J. Jackson Lecture 10-30Electrical & Computer Engineering

    Notch filters

    A notch filter rejects (or passes depending on itsconstruction) frequencies in a pre-defined area(neighborhood) about the center of the frequencyrectangle

    We desire that the filters be zero-phase-shift Must be symmetric about the origin

    A notch with center at (u0,v0) must have a correspondingnotch at (-u0,-v0)

  • 7/29/2019 Low Pass and High Pass Filters

    6/6

    Dr. D. J. Jackson Lecture 10-31Electrical & Computer Engineering

    Notch reject filters

    Notch reject filters are easily constructed asproducts of highpass filters whose centers have

    been translated to the center of the notchesThe general form is:

    Where Hk(u,v) and H-k(u,v) are highpass filterswhose centers are at (uk,vk) and (-uk,-vk)

    Q is the number of notches

    =

    =Q

    kkkNR vuHvuHvuH

    1

    ),(),(),(

    Dr. D. J. Jackson Lecture 10-32Electrical & Computer Engineering

    Notch reject filters (continued)

    The centers at (uk,vk) and (-uk,-vk) are specified withrespect to the center of the frequency rectangle,

    (M/2,N/2) Distances can be calculated as:

    22

    22

    )2/()2/(),(

    )2/()2/(),(

    kkk

    kkk

    vNvuMuvuD

    and

    vNvuMuvuD

    +++=

    +=

    Dr. D. J. Jackson Lecture 10-33Electrical & Computer Engineering

    Notch reject filters (continued)

    A general form for a Butterworth notch reject filterof order n and containing three notch pairs is:

    The constant D0kis the same for each pair ofnotches, but can be different for different pairs

    A notch pass filtercan be expressed as

    +

    +=

    = n

    kkkn

    kk

    NRvuDDvuDD

    vuH2

    0

    3

    12

    0 )],(/[1

    1

    )],(/[1

    1),(

    ),(1),( vuHvuHNRNP

    =