Bi tp ln x l nh

Gii thiu qua ti

ti
7 : Tm hiu v cc php lc s , kho st v lp trnh th nghim cc ng dng ca php
lc trn min tn s vi x l nh mu .
Gio vin hng dn: Nguyn Th Hong Lan
Nhm sinh vin thc hin :
Nguyn Trung Giang
Trn c nh
L Tr Dng

















Li ni u
X l nh ang l mt lnh vc c quan tm v tr thnh mt phn rt quan trng
trng , lin quan n nhiu nghnh khc nh : h thng tin hc , l thuyt thng tin , l thuyt
thng k , tr tu nhn to , nhn dng . y l mt mn khoa hc tng i mi m so vi
nhiu nghnh khoa hc khc . N gm nhiu qu trnh lin tc . u tin l thu nhn nh t
camera , v tinh hay cc b cm ng , Tn hiu ly vo s c s ha thnh tn hiu s v
chuyn sang giai on x l , phn tch hay lu tr li . Trong qu trnh hc , chng em c bit
hng th vi cc phng php lc s x l nh . Chnh v vy nhm em chn ti : " Tm
hiu v cc php lc s , kho st v lp trnh th nghim cc ng dng ca php lc trn min
tn s vi x l nh mu "

y l mt ti rt hay, c bit to nn nhiu nh vi cc mc ch ca ngi dng .
Chng em tm kim v hc hi t c v t ti liu rt nhiu cng vi qu trnh xy dng ng
dng , chng em rt ra c rt nhiu kinh nghim thc t cng nh kin thc su hn v
mn hc , c th tr li c nhng thc mc t lu . Tuy nhin do thi gian lm bi tp ln
khng c nhiu , cng cha c nhiu kinh nghim , ti liu y v lnh vc ny , nn ti
ca chng em kh c th trnh khi thiu xt , nhng cng t c thng li ban u .
Chng em xin cm n c hng dn gip chng em hon thnh bn bo co ny. Em xin
chn thnh cm n .


Nhm sinh vin :
Nguyn Trung Giang 20070911
L Tr Dng 20070553
Trn c nh 20073580









Mc lc

Phn I : Tm hiu v cc php lc s
1. Khi qut v php lc nh
2. Cc b lc s
2.1.nh ngha v m hnh
2.2.Phn loi b lc
- B lc c p ng xung hu hn FIR
- B lc c p ng xung v hn IIR
2.3.Cc b lc s thng dng
- B lc trung bnh
- B lc thng thp
- B lc Laplace
- B lc thng cao

Phn II : Kho st v xy dng ng dng cc php lc trn min tn s
1. C s l thuyt
- Hn ch ca x l trn min khng gian
- tng x l trn min tn s
- Tnh ton chi tit
2. Cc b lc
2.1. Lc thng thp
-Lc tn s thp Idea
- Lc tn s thp Gauss
- Lc tn s thp Butterworth

2.2. Lc thng cao
-Lc thng cao t lc thng
-Lc tn s thp Idea
- Lc tn s thp Gauss
- Lc tn s thp Butterworth

3. Xy dng ng dng











Phn I : Tm hiu v cc php lc s
1. Khi qut v php lc nh
Php lc nh c s dng nhiu trong x l nh , c dng trong gim
nhiu , lm nt nh , cng nh trong pht hin cnh , bin nh Cc php lc nh
ch yu c s dng ngn chn cc tn s cao trong hnh nh , nh lm mn
nh hay tn s thp nh pht hin cnh trong hnh nh . Cc b lc c th chia lm
2 loi theo php ton : lc tuyn tnh v lc phi tuyn. Php lc tuyn tnh l cc
php lc c bn cht l lc tn s nh lc trung bnh, lc thng thp, lc thng cao,
lc o hm. Ngc li cc php lc phi tuyn bao gm lc trung v, lc ng
hnh, lc vi k lng ging gn nht, lc hng r .
Cc php lc nh u s dng cch x l cc b, tc l im nh u ra ch
chu nh hng ca 1 s im nh ln cn theo k thut mt n. Ngi ta cng s
dng php nhn chp ri rc thc hin b lc.

2.1 . nh ngha v m hnh
Mt hnh nh c th c lc trong min tn s hoc trong min khng gian.
Trong k thut lc min khng gian ta s dng mt mt n , t hp im nh t nh
hng ca cc im ln cn. Trong min khng gian ta s dng php nhn chp tn
hiu nh u vo vi b lc s :
Y (m,n ) = H(k,l ) * X(m,n )
Vi K*L << M * N


Hnh 1. B lc 3*3
Ma trn b lc cn c gi l ma trn ht nhn. Cc ma trn ht nhn c th c
nhiu kch thc ty , ph bin nht l ma trn 3*3 (hnh 1) , ngoi ra trong cc
trng hp c th c th s dng cc b lc 5*5 hay 7*7. B lc trong min khng
gian vi ma trn ht nhn kh trc quan v d thc hin. N ph hp vi cm quan
ca chng ta. Tuy nhin cng chnh v kh n gin nn n khng c c s tinh
t. Mt n thng c cc gi tr dng v i xng, nhng khng nht thit phi
nh vy. N c th c chn theo mt phng php no m khng th trc
quan v mt trong cc phng php l lc trn min tn s .
Phng php lc trn min tn s n gin l thc hin cc php bin i
nh trn min tn s. Cc tn hiu u vo, u ra ca nh, cc b lc u c
bin v min tn s.
Y(u,v ) = X(u,v ) * H(u,v)
Vi Y(u,v ) = DFT (Y(m,n));
X(u,v) . H(u,v ) = DFT ( X(m,n ) * H(k,l) )
B lc trn min tn s c 2 loi l b lc thng thp v thng cao. B lc
thng thp thng c s dng lm m nh, gim nhiu,b lc thng cao th
c s dng lm sc nt cnh, bin lm cho nh r nt hn. Thc ra bn cht
ca php lc s u s dng lc tn s , ch l thc hin trn min khng gian hay
min tn s m thi.

Hnh 2 . M hnh b lc
2.2.Phn loi b lc
- B lc c p ng xung hu hn FIR
- B lc c p ng xung v hn IIR

Phn II : Kho st v xy dng ng dng cc php lc trn min tn s

Kt lun



Mean Filter - noise reduction NR using mean of neighborhood

Median Filter - NR using median of neighborhood

Gaussian Smoothing - NR using convolution with a Gaussian smoothing kernel

Conservative Smoothing - NR using maximum and minimum of neighborhood

Crimmins Speckle Removal - more complex NR operator

Frequency Filters - high and low pass image filters, etc

Laplacian/Laplacian of Gaussian Filter - edge detection filter

Unsharp Filter - edge enhancement filter







Mean Filter

Common Names: Mean filtering, Smoothing, Averaging, Box filtering
M t :
Mean filtering l phng php lc kh n gin , trc quan v d thc hin vic lm mn hnh
nh , ngha l gim cng bin i gia mt pixel v pixel k tip . N thng c dng
gim nhiu trong nh .
Nguyn l :
tng ca b lc ny n gin ch l thay th tng gi tr pixel trong nh vi cc gi tr
trung bnh ca cc pixel hng xm, bao gm chnh n. iu ny c tc dng loi b cc gi tr
pixel khng i din cho mi trng xung quanh. Mean filtering thng c coi nh l
convolution filter, chng ging nhau ch cng da trn mt ht nhn i din cho tt c cc
hnh dng kch thc ca c khi c ly mu. Thng th ngi ta s dng ht nhn 3*3 hnh
vung . Ht nhn 5*5 cng c th c s dng v n lm mn nng hn 3*3.




Figure 1 Khi 3*3 thng c s dng trong Mean filter


V d :
Mean filtering c s dng ph bin nht trong vic gim nhiu cho hnh nh , chng ta
ly mt v d minh ha nh sau :




Figure 2a nh gc

Figure 2b nh b nhiu

Figure 2c nh s dng lc 3*3

Figure 2d nh s dng lc 5*5
R rng khi s dng b lc c kch thc 3*3 th nhiu gim i nhng nh mm hn (
hnh 2.c). Nu chng ta tng kch thc b lc ln 5*5 th nh s c rt t nhiu v t cc thnh
phn tn s cao (hnh 2.d) . Ta thy nh trong hnh 2d nh ko gim nhiu nhiu nhng m th
gim i ng k. C th thy 2 vn chnh vi mean filter, l :
- Mt pixel c gi tr khc nhiu vi cc gi tr khc xung quanh s nh hng n tt c
im nh trong khi ca n .
- Vi cc khi cnh, gc b lc s phi ni suy ra gi tr cho cc im nh, t s lm
cho nh b m gc cnh , rt khng tt cho nh c yu cu hnh nh sc nhn u ra.
C 2 vn c th c gii quyt vi medium filter , nhng b li n c thi gian tnh ton
nhanh hn. Tip theo ta s cng xt n medium filter .
Median Filter
Cc b lc trung bnh thng c s dng gim nhiu trong mt hnh nh, phn no
ging nh Mean Filter. Tuy nhin, n gii quyt c vn tt hn.
Nguyn l
Ging nh Mean Filter, Medium Filter xem xt tng im nh trong hnh nh ln lt, v
kim tra cc pixel gn ca n quyt nh n c l i din cho khi nh hay khng. Thay v
ch n gin l thay th cc gi tr pixel vi trung bnh ca cc gi tr pixel ln cn, n thay th
n vi trung bnh ca cc gi tr. Gi tr trung bnh ny c tnh ton bng cch u tin phn
loi tt c cc gi tr im nh t cc khi xung quanh vo s th t v sau thay th cc im
nh c coi l c gi tr pixel trung bnh. (Nu cc khi c xem xt c cha mt s chn cc
pixel, vi mc trung bnh ca hai gi tr pixel trung c s dng.) Hnh 3 m t mt tnh v d.


Hnh 3 . Phng php Medium Filter

Hnh 3 tnh gi tr trung bnh ca mt khi pixel. Nh c th thy, gi tr pixel trung tm
ca 150 l kh khng i din ca cc im nh xung quanh v c thay th bng gi tr trung
bnh: 124. Mt khu ph 3 3 hnh vung c s dng y. Mt khi ln hn s to ra nh c
mn nghim trng hn.
Bng cch tnh ton gi tr trung bnh ca mt khu ph hn l c ngha l b lc, b lc
trung bnh c hai u im chnh trn c ngha l b lc:

* S trung v l trung bnh mnh m hn so vi trung bnh v v vy rt khng i din duy
nht mt im nh trong vng ln cn s khng nh hng n gi tr trung bnh ng k.

* K t khi gi tr trung bnh thc s phi l gi tr ca mt trong nhng im nh trong khu
ph, cc trung bnh lc khng to ra gi tr pixel mi khng thc t khi lc nm gia mt cnh.
V l do ny, cc b lc trung bnh c bo qun tt hn ti cc cnh sc nt hn c ngha l b
lc.
V d :





Figure 3a nh gc

Figure 3b nh b nhiu

Figure 3c nh s dng lc 3*3

Figure 3d nh s dng lc 7*7


Mt trong nhng vn ln vi cc b lc trung bnh l n l tng i tn km v
phc tp tnh ton. tm cc trung bnh n l cn thit sp xp tt c cc gi tr trong khu
ph vo s th t v iu ny l tng i chm, ngay c vi cc thut ton sp xp nhanh nh
quicksort. Cc thut ton c bn c th, tuy nhin, c nng cao phn no tc . Mt k thut
ph bin l thng bo rng khi cc ca s khu ph c trt trn nh, rt nhiu cc im nh
trong ca s u ging nhau t mt trong nhng bc k tip, v tng i t hng ca cc vi
nhau s r rng l khng c thay i. Cc thut ton thng minh s dng cc ny ci thin
hiu sut.

Gaussian Smoothing

Common Names: Gaussian smoothing
Lm mn Gaussian l mt phng php chp 2-D c s dng `m" hnh nh v loi b
chi tit v nhiu. N tng t nh cc Mean Filter, nhng n s dng mt ht nhn khc nhau
m i din cho hnh dng ca mt Gaussian (`hnh chung ') ci bu. Ht nhn ny c mt s
tnh cht c bit c nu chi tit di y.

The Gaussian distribution in 1-D has the form:
Phn phi Gaussian trong mt chiu c dng :

l chun phn phi . Phn phi ny c minh ha nh hnh 4




Figure 4 1-D Gaussian distribution with mean 0 and =1

Trong khng gian 2 chiu phn phi Gauss c dng

Phn phi ny c minh ha nh sau :




Figure 5 2-D Gaussian distribution with mean (0,0) and =1
tng lm mn Gaussian l s dng phn b 2-D l mt chc nng ly lan `im, v iu ny
t c bng cch tch chp. K t khi hnh nh c lu gi nh l mt tp hp cc im nh ring r
chng ta cn phi to ra mt xp x ri rc vi hm Gaussian trc khi chng ti c th thc hin tch
chp. V l thuyt, s phn b Gaussian l khc khng khp mi ni, m s i hi mt ht nhn tch
chp v cng ln, nhng trong thc t n khng c hiu qu hn khong ba lch chun t trung
bnh, v v vy chng ti c th ct cc ht nhn vo thi im ny. Hnh 3 cho thy mt ht nhn
nguyn-ph hp chp c gi tr xp x mt Gaussian vi mt l 1.0.




Figure 3 Discrete approximation to Gaussian function with =1.0

Once a suitable kernel has been calculated, then the Gaussian smoothing can be performed using
standard convolution methods. The convolution can in fact be performed fairly quickly since the
equation for the 2-D isotropic Gaussian shown above is separable into x and y components. Thus
the 2-D convolution can be performed by first convolving with a 1-D Gaussian in the x direction,
and then convolving with another 1-D Gaussian in the y direction. (The Gaussian is in fact the
only completely circularly symmetric operator which can be decomposed in such a way.) Figure
4 shows the 1-D x component kernel that would be used to produce the full kernel shown in
Figure 3 (after scaling by 273, rounding and truncating one row of pixels around the boundary
because they mostly have the value 0. This reduces the 7x7 matrix to the 5x5 shown above.). The
y component is exactly the same but is oriented vertically.
Khi mt ht nhn ph hp c tnh ton, sau l lm mn Gauss c th c thc hin
bng cch s dng cc phng php chp tiu chun. chp Cc trn thc t c th c thc hin
kh nhanh chng k t khi phng trnh cho 2-D ng hng Gaussian hin th trn l tch
thnh cc thnh phn x v y. Do , chp 2-D c th c thc hin bng cch u tin
convolving vi mt Gaussian 1-D theo hng x, v sau convolving vi mt Gaussian 1-D
theo hng y. (Gaussian l trong thc t, cc nh iu hnh ch hon ton trn i xng c th
c phn hy theo cch nh vy.) Hnh 4 cho thy 1-D x thnh phn ht nhn m c th c
s dng sn xut ht nhn ton th hin trong hnh 3 (sau khi m rng quy m ca 273 , lm
trn s v ct xn mt hng cc im nh xung quanh ranh gii, v h ch yu c gi tr 0 iu
ny lm gim. cc ma trn 7x7 n 5x5 hin th trn.). Cc thnh phn y l chnh xc ging
nhau nhng c nh hng theo chiu dc.




Figure 4 One of the pair of 1-D convolution kernels used to calculate the full kernel shown in Figure 3
more quickly.

A further way to compute a Gaussian smoothing with a large standard deviation is to convolve
an image several times with a smaller Gaussian. While this is computationally complex, it can
have applicability if the processing is carried out using a hardware pipeline.
The Gaussian filter not only has utility in engineering applications. It is also attracting attention
from computational biologists because it has been attributed with some amount of biological
plausibility, e.g. some cells in the visual pathways of the brain often have an approximately
Gaussian response.
Mt cch khc tnh ton mt mn Gaussian vi mt lch chun ln l cy dy leo mt hnh
nh nhiu ln vi mt Gaussian nh hn. Trong khi iu ny c tnh ton phc tp, n c th
c ng dng nu ch bin c thc hin bng cch s dng mt ng ng dn phn cng.

Cc b lc Gaussian khng ch c ch trong cc ng dng k thut. N cng thu ht s ch t
cc nh sinh hc tnh ton bi v n c c l do c mt s lng s tht sinh hc, v d: mt s
t bo trong con ng th gic ca no b thng c mt phn ng khong Gaussian.

Guidelines for Use
The effect of Gaussian smoothing is to blur an image, in a similar fashion to the mean filter. The
degree of smoothing is determined by the standard deviation of the Gaussian. (Larger standard
deviation Gaussians, of course, require larger convolution kernels in order to be accurately
represented.)
The Gaussian outputs a `weighted average' of each pixel's neighborhood, with the average
weighted more towards the value of the central pixels. This is in contrast to the mean filter's
uniformly weighted average. Because of this, a Gaussian provides gentler smoothing and
preserves edges better than a similarly sized mean filter.
Hiu qu ca vic lm mn Gaussian blur l mt hnh nh, trong mt thi trang tng t nh c
ngha l b lc. Mc lm mn c xc nh bi lch chun ca cc Gaussian. (Ln hn
lch chun Gaussians, tt nhin, yu cu nhn chp ln hn c chnh xc i din.)

Gaussian Cc kt qu trung bnh `trng 'ca vng ln cn ca mi im nh, vi bnh qun nhiu
hn i vi gi tr ca im nh trung tm. iu ny tri ngc vi mc trung bnh c ngha l
b lc trng ca thng nht. Bi v iu ny, Gaussian mt cung cp lm mn nh nhng v bo
ton cc cnh tt hn so vi mt kch thc tng t nh vy c ngha l b lc.

One of the principle justifications for using the Gaussian as a smoothing filter is due to its
frequency response. Most convolution-based smoothing filters act as lowpass frequency filters.
This means that their effect is to remove high spatial frequency components from an image. The
frequency response of a convolution filter, i.e. its effect on different spatial frequencies, can be
seen by taking the Fourier transform of the filter. Figure 5 shows the frequency responses of a 1-
D mean filter with width 5 and also of a Gaussian filter with = 3.
Mt trong nhng lun c nguyn tc s dng Gaussian nh mt b lc lm mn l do p ng
tn s ca n. Hu ht cc b lc da trn chp lm mn lm lowpass b lc tn s. iu ny c
ngha l hiu ng ca h l loi b cc thnh phn tn s cao khng gian t mt hnh nh. Cc
p ng tn s ca tch chp mt b lc, ngha l hiu ng ca n trn tn s khc nhau v khng
gian, c th c xem bng cch bin i Fourier ca cc b lc. Hnh 5 cho thy cc phn ng
tn s ca mt 1-D c ngha l b lc vi rng 5 v cng l ca mt b lc Gaussian vi eqn:
eqnsigma = 3.




Figure 5 Frequency responses of Box (i.e. mean) filter (width 5 pixels) and Gaussian filter ( = 3 pixels).
The spatial frequency axis is marked in cycles per pixel, and hence no value above 0.5 has a real
meaning.

Both filters attenuate high frequencies more than low frequencies, but the mean filter exhibits
oscillations in its frequency response. The Gaussian on the other hand shows no oscillations. In
fact, the shape of the frequency response curve is itself (half a) Gaussian. So by choosing an
appropriately sized Gaussian filter we can be fairly confident about what range of spatial
frequencies are still present in the image after filtering, which is not the case of the mean filter.
This has consequences for some edge detection techniques, as mentioned in the section on zero
crossings. (The Gaussian filter also turns out to be very similar to the optimal smoothing filter
for edge detection under the criteria used to derive the Canny edge detector.)

Hnh 5 Tn s p ng ca Box (tc l trung bnh) b lc (rng 5 im nh) v b lc Gaussian
(eqn: eqnsigma = 3 im). Cc trc tn s khng gian c nh du theo chu k mi im nh,
v do khng c gi tr trn 0,5 c mt ngha thc s.


C hai b lc suy nhc tn s cao hn tn s thp, nhng c ngha l cuc trin lm b lc p
ng tn s dao ng ca n. Cc Gaussian mt khc cho thy khng c dao ng. Trong thc t,
hnh dng ca ng cong p ng tn s l chnh n (na) Gaussian. V vy, bng cch chn
mt cch thch hp c kch thc b lc Gaussian chng ta c th kh t tin v nhng g lot cc
tn s khng gian vn cn hin din trong hnh nh sau khi lc, m khng phi l trng hp c
ngha l b lc. iu ny c hu qu i vi mt s k thut pht hin bin, nh cp
phn trn khng qua li. (Cc b lc Gaussian cng ha ra l rt tng t nh cc b lc lm mn
ti u pht hin cnh theo cc tiu chun c s dng ly c pht hin cnh khn
ngoan.)