DIP3E Chapter03 Art

Digital Image Processing, 3rd ed.
Gonzalez & Woods

www.ImageProcessingPlace.com
Chapter 3
Intensity Transformations & Spatial
Filtering
Chapter 3: Intensity transformations and spatial filtering
Generic form:
g(x, y) = T [f (x, y)]
new intensity value at (x,y) is a function of the

old intensity value at
(x,y), or Sxy
© 1992–2008 R. C. Gonzalez & R. E. Woods

Gonzalez & Woods
Chapter 3
Intensity Transformations & Spatial Filtering
Example of spatial filtering:

blurring by 3x3 neighborhood
Create a 3x3 mask, and run it over the whole image to generate the new
image (can also use conv2)
Here I have
run the mask
only on the
left half of
the
picture.
© 1992–2008 R. C. Gonzalez & R. E. W ood
s
Tuesday, February 1, 2011
Gonzalez & Woods
Chapter 3
Example of spatial filtering:

blurring by 3x3 neighborhood
Create a 3x3 mask, and run it over the whole image to generate the new
image (can also use conv2)
Here I have
run the mask
Example of a blurred
only on the
image by a 3x3
left half of mask
the picture.
© 1992–2008 R. C. Gonzalez & R. E. W ood
s
Gonzalez & Woods
Chapter 3
Filtering
Simplest example:
1-pixel neighborhood (the pixel itself), in which case we operate
directly on the intensity of the pixel.
This is called point processing as opposed to

neighborhood processing.
Typical operations we will perform:

Image enhancement
contrast stretching
Is the task of manipulating an image so
thresholding that the result is more suitable for specific
applications.
Mostly visual (subjective), no general
quantitative assessment.
Gonzalez & Woods
Chapter 3
Filtering
Contrast stretching (sigmoid):
low intensity areas are mapped

to even lower
high intensity areas are mapped
to even higher
medium intensity areas are
“stretched”.

Gonzalez & Woods
Chapter 3
Filtering

to even lower
to even higher
“stretched”.
© 1992–2008 R. C. Gonzalez & R. E. Woods Original contrast stretched

Gonzalez & Woods
Chapter 3
Filtering

to even lower
to even higher
“stretched”.
© 1992–2008 R. C. Gonzalez & R. E. Woods Original contrast stretched

Gonzalez & Woods
Chapter 3
Thresholding: we choose a threshold parameter k: intensities below

this parameters are set to 0, those above are set to 1.
This produces a binary image.

Gonzalez & Woods
Chapter 3
Thresholding: we choose a threshold parameter k: intensities below

this parameters are set to 0, those above are set to 1.
This produces a binary image.
Thresholded image (setting to 1 the

values above 150, 0 the others)

Gonzalez & Woods
Chapter 3
Other maps of the type s=T(r)

Negative map: s = (L − 1) − r
useful when one is interested in gray/white areas and the black areas
are dominating

Gonzalez & Woods
Chapter 3

are dominating

Gonzalez & Woods
Chapter 3

are dominating

Gonzalez & Woods
Chapter 3

are dominating

Gonzalez & Woods
Chapter 3
Log map: s = c log(1 + r)

maps lower intensities to a
wider range, while white
intensities are
mapped to a smaller range

Gonzalez & Woods
Chapter 3
Log map: s = c log(1 + r)

maps lower intensities to a
wider range, while white
intensities are
mapped to a smaller range

Gonzalez & Woods
Chapter 3
Filtering
Power-law (gamma) map:

s = cr γ
These subsume the log and

inverse log, but are more
powerful, as just varying the
gamma one gets a whole family
of transformations.

Gonzalez & Woods
Chapter 3
Filtering
Power-law (gamma) map: gamma=0.2
s = cr γ

of transformations.

Gonzalez & Woods
Chapter 3
Filtering
Power-law (gamma) map: gamma=0.2
s = cr γ

of transformations.
gamma=3

Gonzalez & Woods
Chapter 3
Filtering

Gonzalez & Woods
Chapter 3
Filtering
Piecewise linear transformations:
Contrast stretching by
piecewise linear transformations.
(r1, s1)
(r2, s2)
Control points

Gonzalez & Woods
Chapter 3
Filtering

Gonzalez & Woods
Chapter 3
r1=0=s1
r2=L-1=s2
Identity

Gonzalez & Woods
Chapter 3
Filtering

Gonzalez & Woods
Chapter 3
r1=r2
s1=0, s2=L-1
Thresholding

Gonzalez & Woods
Chapter 3
r1=r2
s1=0, s2=L-1
Thresholding
Other combinations give different effects on the image.

Gonzalez & Woods
Chapter 3
Other combinations give different effects on the image.

Gonzalez & Woods
Chapter 3
Intensity level slicing:

to highlight a specific range of intensities.
Most popularly implemented as:

showing white in one range and black in the rest
showing white in one range and graytone in the rest

Gonzalez & Woods
Chapter 3
Filtering

Gonzalez & Woods
Chapter 3
Bit slicing:
2^7 2^6 2^5 2^4 2^3 2^2 2^1 2^0
0 1 0 1 0 0 1 0
82=64+16+2

Gonzalez & Woods
Chapter 3
Filtering
The reconstructed image is obtained as
2^0*bit_plane1 + 2^1*bit_plane2 + …. + 2^7*bit_plane8
bit plane 1
The most
significant planes
are often those with
high bit number
bit plane 8
Gonzalez & Woods
Chapter 3
Filtering

Gonzalez & Woods
Chapter 3
Filtering

Gonzalez & Woods
Chapter 3
Filtering
Histogram processing
Histogram: is a discrete function that counts how many pixels have a

given intensity value.
h(r k ) = nk Number of pixels that have such

intensity value
k-th intensity value

Gonzalez & Woods
Chapter 3
Filtering
Histogram processing
Histogram: is a discrete function that counts how many pixels have a

given intensity value.
h(r k ) = nk Number of pixels that have such

intensity value
k-th intensity value

Gonzalez & Woods
Chapter 3
It is typical to scale nk by MN, the number of pixels. The corresponding number represents the
probability of a pixel having the given intensity.
nk
p(r k ) = p(r k ) = 1
MN k
nk = M N
k Probability that a random pixels has intensity r k .

Gonzalez & Woods
Chapter 3
Filtering

Gonzalez & Woods
Chapter 3
Filtering
It is typical to scale nk by MN, the number of pixels. The corresponding number represents the probability of a
pixel having the given intensity.
nk
∑ p ( r k )=1
∑ nk= MN p ( r k ) =
MN k
k
Probability that a random pixels has intensity rk.

Gonzalez & Woods
Chapter 3
Image enhancement by histogram equalization
Consider an intensity transformation s = ( r input intensity level, s

output intensity level). Let us assume that
• T is monotonically increasing that is T

(r1) ≤ T (r2) for r1≤r2
(strictly increasing if T −1 is required)
• T maps [0,L − 1] to [0,L − 1]

Gonzalez & Woods
Chapter 3
We can identify the intensity level to a random variable in [0,L − 1].

Then the function p = h/MN is the probability density function of such
random variable.
Denote by
•pr the probability distribution of the original image (r-variable) and
•ps the probability distribution of the image with intensity s = . These
two distributions will generally be diﬀerent.
Thus, given the change of variables s = , the relation between ps and

pr is given by the formula
𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) | |
𝑑𝑟
𝑑𝑠
Gonzalez & Woods
Chapter 3
𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 )| |
𝑑𝑟
𝑑𝑠
Where does it come from?
ps (s)
p r (r) (Recall, probability
distributions are
positive)
dr ds
r s
s=
The gray regions must have the same area (we are just relabeling the variables)
𝒑 𝒓 ( 𝒓 ) 𝒅𝒓 =𝒑 𝒔 ( 𝒔 ) 𝒅𝒔
(take the absolute value is for the case when T is decreasing)
Gonzalez & Woods
Chapter 3
| | 𝑑𝑟 1
𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) 𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) ′ (𝑟)
𝑑𝑠 𝑇
(Alternative formulation)
Example. Consider the transformation

𝒓
s = 𝑻 ( 𝒓 ) = ( L − 1 ) ∫ 𝝆𝒓 ( 𝝎 ) 𝒅 𝝎
𝟎
The transformation is strictly increasing because pr is a positive function.

Gonzalez & Woods
Chapter 3
| |𝑑𝑟 1
𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) 𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) ′ (𝑟)
𝑑𝑠 𝑇
Examp l e. Consider the transformation

𝒓
s = 𝑻 ( 𝒓 ) = ( L −. 1 ) ∫ 𝝆𝒓 ( 𝝎 ) 𝒅 𝝎
𝟎
Cumulative probability
Gonzalez & Woods
Chapter 3
| |𝑑𝑟 1
𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) 𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) ′ (𝑟)
𝑑𝑠 𝑇
Examp l e. Consider the transformation

𝒓
s = 𝑻 ( 𝒓 ) = ( L −. 1 ) ∫ 𝝆 𝒓 ( 𝝎 ) 𝒅 𝝎
𝟎
Cumulative probability
Scaling factor so that T(L-1) = L-1
Gonzalez & Woods
Chapter 3
What is 𝒑 𝒔 ( 𝒔 ) ?
1
Let’s use the formula 𝑝 𝑠 ( 𝑠 )=𝑝𝑟 ( 𝑟 ) ′ (𝑟)
𝑇
We derive with respect to
′ 𝑑
𝑇 (𝑟 )= 𝑇 ( 𝑟 )=( 𝐿 − 1) 𝑝 𝑟 ( 𝑟 )
𝑑𝑟
Hence
1 1 1 The uniform
𝑝 𝑠 ( 𝑠 )= 𝑝𝑟 ( 𝑟 ) ′ =𝑝 𝑟 (𝑟 ) =
|𝑇 (𝑟 )| |( 𝐿 −1)𝑝 𝑟 (𝑟 )| 𝐿 −1
distribution!
Gonzalez & Woods
Chapter 3
For discrete density functions, the equivalent of

𝒓
s = 𝑻 ( 𝒓 ) = ( L − 1 ) ∫ 𝝆𝒓 ( 𝝎 ) 𝒅 𝝎
𝟎
is obtained by replacing the integral by a sum,

𝐤
𝒔 𝒌 =𝑻 (𝒓 𝒌 ) = ( L − 1) ∑ 𝐩 𝐫 (𝐫 𝐣)
𝐣=𝟎
𝐤
(L − 1)
=
𝐌𝐍 𝐣=𝟎
∑ 𝐧𝐣 k = 0 , …, L − 1
This transformation is called histogram equalization/linearization.

Gonzalez & Woods
Chapter 3
Example:
𝐤
𝒔 𝒌 =𝑻 (𝒓 𝒌 ) = ( L − 1) ∑ 𝐩 𝐫 (𝐫 𝐣)
𝐣=𝟎
𝐤
(L − 1)
=
𝐌𝐍 𝐣=𝟎
∑ 𝐧𝐣 k = 0 , …, L − 1
Gonzalez & Woods
Chapter 3
Example:
k
s k =T (r k ) = ( L − 1) ∑ p r (r j )
j=0
k
(L − 1)
=
MN j=0
∑ nj k = 0 , … , L − 1
s0 = (L − 1)pr(r0) = 7 · 0.19 = 1.33
s1 = 7 · (0.19 + 0.25) = 3.08
s3 = 7 · (0.19 + 0.25 + 0.21) = 4.55

Gonzalez & Woods
Chapter 3
Example:
k
s k =T (r k ) = ( L − 1) ∑ p r (r j )
j=0
k
(L − 1)
=
MN j=0
∑ nj k = 0 , … , L − 1
s0 = (L − 1)pr(r0) = 7 · 0.19 = 1.33 s0 = 1
s1 = 3
s1 = 7 · (0.19 + 0.25) = 3.08
s2 = 5
s3 = 7 · (0.19 + 0.25 + 0.21) = 4.55
and so on...
Gonzalez & Woods
Chapter 3
Example:
k
s k =T (r k ) = ( L − 1) ∑ p r (r j )
j=0
k
(L − 1)
=
MN j=0
∑ nj k = 0 , … , L − 1
s0 = (L − 1)pr(r0) = 7 · 0.19 = 1.33 s0 = 1
s1 = 3
s1 = 7 · (0.19 + 0.25) = 3.08
s2 = 5
s3 = 7 · (0.19 + 0.25 + 0.21) = 4.55
and so on...
Since the sk is an intensity transformation, it must be an integer
value. This rounding causes the resulting histogram not necessarily
© 1992–2008. R.C. Gonzalez & R. E. Woods to result into a uniform histogram.
Gonzalez & Woods
Chapter 3
Though the resulting histogram is not uniform, the final result is that
the intensity values are spread out (span a wider range).
This results in an image that has a enhanced contrast.

Gonzalez & Woods
Chapter 3
Though the resulting histogram is not uniform, the final result is that
the intensity values are spread out (span a wider range).
This results in an image that has a enhanced contrast.
NB. Note that the resulting image might reduce the actual number of intensity
levels (because of discretization).
Gonzalez & Woods
Chapter 3
Seljord lake
(Norwegian equivalent
of Loch Ness)
Picture taken with a cell

phone camera (low contrast)
Gonzalez & Woods
Chapter 3
Seljord lake
of Loch Ness)

Before and after

histogram equalization
Gonzalez & Woods
Chapter 3
Seljord lake
of Loch Ness)

Before and after

histogram equalization
Gonzalez & Woods
Chapter 3
Matlab command
for histogram Faces
equalization: on
Mars
histeq
Gonzalez & Woods
Chapter 3
Not in all cases

contrast
enhancement
gives a visual
enhancement
of the picture
(Enhancement is
a visual property,
hence partly
subjective).
For landscapes, it works
generally
well. For portraits,
depends
Gonzalez & Woods
Chapter 3
dark image
light image
low contrast
high contrast
Gonzalez & Woods
Chapter 3
Histogram matching
As we have seen, the histogram equalization does not necessarily

“enhance” equally well all the images.
Example: the face
Assume that we have found the “winning formula”, that is, a

histogram that works well for enhancing a given type of images.
We might wish to apply an intensity transformation to match the

prescribed histogram (probability density distribution).
Gonzalez & Woods
Chapter 3
pr(r) pz(z)
Gonzalez & Woods
Chapter 3
equidistribution
pr(r) pz(z)
Gonzalez & Woods
Chapter 3
equidistribution
T (r)
pr(r) pz(z)
Gonzalez & Woods
Chapter 3
equidistribution
T (r) G(z)
pr(r) pz(z)
Gonzalez & Woods
Chapter 3
equidistribution
T (r) G(z) G−1(s)
pr(r) pz(z)
Gonzalez & Woods
Chapter 3
equidistribution
T (r) G(z) G−1(s)
pr(r) pz(z)
z = G−1 (s) = G−1 (T (r))
Therefore, to match the histograms we have to compute the

equidistribution of the probability density of first image and then the
inverse using the equidistribution of the probability density of the
given histogram.
Gonzalez & Woods
Chapter 3
equidistribution
T (r) G(z) G−1(s)
pr(r) pz(z)
z = G−1 (s) = G−1 (T (r))
In practice, this requires the computation of the equalization of

the first image and then G−1
For discrete probabilities, it is ``exactly’’ the same.

Gonzalez & Woods
Chapter 3
Example:
k
s k =T (r k ) = ( L − 1) ∑ p r (r j )
j=0 G(z0) = 0(0)
s0 = 1
G(z1) = 0(0)
k s1 = 3 (L-1)*cum.prob.
(L − 1)
=
MN j=0
∑ n j k = 0 , … , L − 1
s2 = 5 G(z2) = 0(0)
=7*0.15=1.05
(L-1)*cum.prob.
=7*(0.15+0.2)
s3 = 6 G(z3) = 1(1.05) =2.45
G(z4) = 2(2.45)
s4 = 6
G(z5) = 5(4.55)
s5 = 7
G(z6) = 6(5.95)
s6 = 7
s7 = 7 G(z7) = 7(7)
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)
Now we can compute the inverse of G for the range of interest.

Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7)
= 7(7)
s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
G(z0) = 0(0)
G(z1) = 0(0)
G(z2) = 0(0)
G(z3) = 1(1.05)
G(z4) = 2(2.45)
G(z5) = 5(4.55)
G(z6) = 6(5.95)
G(z7) = 7(7)

s0 = 1
s1 = 3
s2 = 5
s3 = 6
s4 = 6
s5 = 7
s6 = 7
s7 = 7
Gonzalez & Woods
Chapter 3
Since we are working with integer number, because of rounding

the computed transformation does not give an exact histogram
match.
However, there will be a tendency to move the intensity values to

have a similar match.
Gonzalez & Woods
Chapter 3
Pollen image Pollen image

(reference img)
Gonzalez & Woods
Chapter 3

(reference img)
Gonzalez & Woods
Chapter 3

(reference img)
Gonzalez & Woods
Chapter 3
How to do the histogram matching in Matlab?
J=histeq(I) : histogram equalization of the image I
J=histeq(I, HIST) : creates J from I by intensity

transformation so that the histogram of J and HIST match.
Gonzalez & Woods
Chapter 3
The histogram transformations that we have seen up to this point are

global, in the sense that they act on the whole image.
In some applications, one might wish to apply different

transformations to different regions of the image
Local histogram processing

Gonzalez & Woods
Chapter 3
Further image statistics (normalized quantities):
Definitions:
Average intensity:
𝐿 −1
m = ∑ 𝑟 𝑖 𝑝(𝑟 𝑖 ) (also called mean intensity)
𝑖=1
nth moment:
𝐿 −1
𝜇𝑛 = ∑ (𝑟 𝑖 −𝑚) 𝑝(𝑟 𝑖 )
𝑛
𝑖=1
Second moment:
(this is the intensity
𝐿 −1
𝜇 2 = ∑ (𝑟 𝑖 − 𝑚)2 𝑝(𝑟 𝑖 ) variance, and equals
𝑖=1
Gonzalez & Woods
Chapter 3
The mean m (average) and the variance can be computed directly from the
image data,
𝑀 −1 𝑁 −1
1
m= ∑
𝑀𝑁 𝑥=0
∑ 𝑓 (𝑥 , 𝑦)
𝑦 =0
𝑀 −1 𝑁 −1
1
𝜎=2
𝑀𝑁
∑ ∑ ( 𝑓 ( 𝑥 , 𝑦 ) − 𝑚) 2
𝑥=0 𝑦 =0
It is an easy exercise to see that the values computed in this manner

are the same obtained with the probability density function.
Gonzalez & Woods
Chapter 3
The previous definitions of mean, average, moments,

can easily be adapted to a subregion of an image (instead of the full
one).
𝑀 −1
1 1
𝑚𝑠 𝑥𝑦 =
𝑀𝑁
∑ 𝑟 𝑖 𝑝𝑠 𝑥𝑦 ( 𝑟 𝑖 ) = ∑
|𝑆 𝑥𝑦| (𝑠,𝑡)∈𝑆
𝑓 (𝑠,𝑡)
𝑥=0 𝑥𝑦
𝐿−1
1
𝜎 = ∑ (𝑟 𝑖 −𝑚𝑠 𝑥𝑦 ) 𝑝𝑠 𝑥𝑦 ( 𝑟 𝑖 )= ∑
2 2 2
( 𝑓 ( 𝑠,𝑡 ) −𝑚𝑠 𝑥𝑦 )
𝑖=1 |𝑆 𝑥𝑦|(𝑠 ,𝑡)∈𝑆 𝑥𝑦
Gonzalez & Woods
Chapter 3
Example: we have an image, we are satisfied from parts of it

(foregreound), want to enhance other parts (background)
Gonzalez & Woods
Chapter 3

Gonzalez & Woods
Chapter 3
Gonzalez & Woods
Chapter 3
Filtering
Fundaments of spatial filtering
Filtering = sieving, removing unwanted components
Image filtering adapts and expands

techniques already used in signal
processing.
Filtering is accomplished directly

on the image using masks (kernels,
templates, windows)

Gonzalez & Woods
Chapter 3
Filtering
We have already seen an example of spatial filtering
(blurring by 9 pixels averaging (D8))
This is a special example

of linear filter
as the result (response) is
obtained as a linear
combination of the input
image pixel values.

Gonzalez & Woods
Chapter 3
Filtering
We have already seen an example of spatial filtering
(blurring by 9 pixels averaging (D8))
This is a special example

of linear filter
as the result (response) is
obtained as a linear Example of a blurred
image by a 3x3
combination of the input mask
image pixel values.

Gonzalez & Woods
Chapter 3
A generic D 8 linear filter will have the following expression

g(x, y) = w(−1, −1)f (x − 1,y − 1) + w(−1, 0)f (x − 1, 0) +
·· ·
+ w(0, 0)f (x, y) + ·· · + w(1, 1)f (x + 1,y +
or, for a generic mask using
1) mn pixels,
𝑎 𝑏
g( x ,y) = ∑ ∑ 𝑤(𝑠,𝑡) 𝑓 (𝑥+𝑠 , 𝑦+1)
𝑠=− 𝑎 𝑡=− 𝑏
where
m = 2a + n = 2b + 1
1,

Gonzalez & Woods
Chapter 3
Filtering
2 important filter concepts: Spatial correlation and convolution
Correlation: is the process of moving a filter mask over an image and

computing the sum of the (pointwise) product.
𝑎 𝑏
g( x ,y) = ∑ ∑ 𝑤(𝑠,𝑡) 𝑓 (𝑥+𝑠 , 𝑦+1)
𝑠=− 𝑎 𝑡=− 𝑏
Convolution: mechanically the same (except

for the mask, rotated by 180 degrees).
𝑎 𝑏
g( x ,y) = ∑ ∑ 𝑤(𝑠,𝑡) 𝑓 (𝑥 − 𝑠, 𝑦 −1)
𝑠=− 𝑎 𝑡=− 𝑏
© 1992–2008 R. C. Gonzalez & R. E. W oods

DIP3E Chapter03 Art

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

DIP3E Chapter03 Art

Uploaded by

Copyright:

Available Formats

Digital Image Processing, 3rd ed.

Gonzalez & Woods

new intensity value at (x,y) is a function of the

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

Example of spatial filtering:

Example of spatial filtering:

This is called point processing as opposed to

Typical operations we will perform:

Contrast stretching (sigmoid):

low intensity areas are mapped

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

Tuesday, February 1, 2011

Contrast stretching (sigmoid):

low intensity areas are mapped

© 1992–2008 R. C. Gonzalez & R. E. Woods Original contrast stretched

Contrast stretching (sigmoid):

low intensity areas are mapped

© 1992–2008 R. C. Gonzalez & R. E. Woods Original contrast stretched

Thresholding: we choose a threshold parameter k: intensities below

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

Tuesday, February 1, 2011

Thresholding: we choose a threshold parameter k: intensities below

Thresholded image (setting to 1 the

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

Tuesday, February 1, 2011

Other maps of the type s=T(r)

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

Tuesday, February 1, 2011

Other maps of the type s=T(r)

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

Tuesday, February 1, 2011

Other maps of the type s=T(r)

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

Tuesday, February 1, 2011

Other maps of the type s=T(r)

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

Tuesday, February 1, 2011

Log map: s = c log(1 + r)

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

Tuesday, February 1, 2011

Log map: s = c log(1 + r)

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

Tuesday, February 1, 2011

Power-law (gamma) map:

These subsume the log and

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

Tuesday, February 1, 2011

Power-law (gamma) map: gamma=0.2

These subsume the log and

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

Tuesday, February 1, 2011

Power-law (gamma) map: gamma=0.2

These subsume the log and

Tuesday, February 1, 2011

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

Tuesday, February 1, 2011

Piecewise linear transformations:

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

Tuesday, February 1, 2011

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

Tuesday, February 1, 2011

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

Tuesday, February 1, 2011