Image Acquisition & Digitization

Human and Computer Vision

• We can’t think of image processing without
considering the human vision system. We
observe and evaluate the images that we
process with our visual system.
• Without taking this elementary fact into
consideration, we may be much misled in
the interpretation of images.

4/3/2025 (#/62) 2
 Simple questions
• What intensity differences can we
distinguish?
• What is the spatial resolution of our eye?
• How accurately we estimate and compare
distances and areas?
• How do we sense colors?
• By which features can we detect and
distinguish objects?
4/3/2025 (#/62) 3
 Test images a) Parallel lines
with up to 5%
Distances and areas estimations
difference in
length.

b) Circles with
up to 10%
difference in
radius.
c) The vertical
line appears
longer but
actually has the
same length as
the horizontal
line.
d) Deception by perspective: the upper line appears longer
4/3/2025 (#/62) 4
than the lower one but actually same length.
 Brightness
• Intensity and brightness are two different
phenomena.
• intensity is a physical property and can be
measured.
• brightness is a psycho-visual concept and may
be described as the sensation to the light
intensity and not just simple function of intensity.
• The contrast may be defined as the difference in
perceived brightness.

4/3/2025 (#/62) 5
 Mach band pattern
Is it the same level of
darkness around area D
and B ?
The brightness pattern
perceived is a darker stripe
in region D and a brighter
one in the region B whereas
actually the region from D to
B has the same intensity.
Note: it was named for
Ernst Mach who
discovered
4/3/2025
the (#/62) 6
phenomenon in 1865.
 Mach band pattern

Mach bands
4/3/2025 (#/62) 7
 Simultaneous contrast
• contrast sensitivity of the eye refers to the ratio of
the difference in brightness B1 (of an object) and
brightness B (of its immediate surrounding) to B.
• Which small square is the darkest one ?
All the small squares have
exactly the same intensity, but
they appear to the eye
progressively darker as the
background becomes brighter.
Region’s perceived brightness
4/3/2025 does
(#/62)not depend simply on its
8
 Human Perception Phenomena
Eye fills non-existing
information
Or
wrongly perceives
geometrical properties of
the object

visual illusions

4/3/2025 (#/62) 9
 Example of Digital Image Acquisition Process

4/3/2025 (#/62) 10
 Image Formation
• For natural images we need a light • Light reflects from a
source ( λ:wavelength of the source): point and the reflected
E (x; y; z; λ): incident light on a light is captured by an
point (x; y; z) world coordinates of imaging device c (x; y; z;
the point λ) = E(x; y; z; λ) × r(x; y;
• Each point in the scene has a z; λ)
reflectivity function: r (x; y; z; λ)

4/3/2025 (#/62) 11
 Inside the Camera - Projection

Projection (P) from world coordinates (x; y; z) to camera or image

coordinates (x’; y’) [cp (x’; y’; λ) = P(c(x; y; z; λ))].

4/3/2025 (#/62) 12
 Projections
• There are two types of projections (P) of interest to us:
– 1. Perspective Projection
• Objects closer to the capture device appear bigger. Most
image formation situations can be considered to be under
this category, including images taken by camera and the
human eye.
– 2. Orthographic Projection
• This is “unnatural”. Objects appear the same size
regardless of their distance to the “capture device”.

• Both types of projections can be represented via

mathematical formulas.
4/3/2025 (#/62) 13
 Example - Perspective

• Perspective Projection: 1= 2 l1 < l2 → δ2< δ1

4/3/2025 (#/62) 14
 Example - Orthographic

• Perspective Projection: 1= 2 l1 < l2 → δ2= δ1

4/3/2025 (#/62) 15
 Inside the Camera - Sensitivity

• We have cp (x’; y’; λ) ; V(λ) is the sensitivity function of a

capture device. Each capture device has such a function which
determines how sensitive it is in capturing the range of
wavelengths (λ) present in cp (x’; y’; λ).

The result is an “image function” which determines the amount of

reflected light that is captured at the camera coordinates (x’; y’).
f(x’; y’) = ∫cp (x’; y’; λ)V(λ) dλ (#/62)
4/3/2025 16
 Sensitivity and Color
• For a camera that captures color images, imagine that it has
three sensors at each (x0; y0) with sensitivity functions tuned
to the colors or wavelengths red, green and blue, outputting
three image functions:
• fR (x0; y0) = ∫cp (x0; y0; λ)VR (λ) dλ
• fG (x0; y0) = ∫cp (x0; y0; λ)VG (λ) dλ
• fB (x0; y0) = ∫cp (x0; y0; λ)VB (λ) dλ

4/3/2025 (#/62) 17
 Digital Image Formation

• The image function fc (x,y) is still a function of:

– x in [xmin, xmax] and y in [ymin, ymax]
– a continuous function in the respective intervals.
• The values taken by the image function fc (x,y) are real numbers
which again continuous in the interval fc (x,y) in [fmin, fmax].
• Digital computers cannot process parameter/function which is
continuous.
• We have to discretize:
– 1. x, y → xi , yj (i = 0,…, N-1) (j = 0,…, M -1): Sampling

– 2. fc (xi , yj) → fd (xi , yj): Quantization.

4/3/2025 (#/62) 18
 Sampling

4/3/2025 (#/62) 19
 Sampling (Contd.)

4/3/2025 (#/62) 20
 Sampling

Sampling

4/3/2025 (#/62) 21
 Quantization
max max

min Quantization min

max max

min Sampling min

4/3/2025 (#/62) 22
 Generating a digital image

a) Continuous image
b) Intensity value
along the scan line
b ab
c) Sampling and
a quantization
d) Digital scan line

c d
4/3/2025 (#/62) 23
 Quantization
• fc (i; j) (i = 0,…, N-1; j = 0,…, M -1). We have the
second step of discretization left.
• fc (i; j) in [fmin, fmax]; for all i and j.
• Discretize the values fc (i; j) to P levels (P=2t) as
follows:
– Let Q = (fmax - fmin)/P .
• fq (i; j) = Q(fc (i; j))
– where Q(fc (i; j)) = (k + 1/2) Q + fmin
– if and only if fmin + k Q ≤ fc (i; j) < fmin + (k + 1) Q
• for k = 0,…, P-1

4/3/2025 (#/62) 24
 Quantization (Contd.)

4/3/2025 (#/62) 25
 Images as Matrices
• A digital image can be considered a matrix whose row and
column indices identify a point in the image and the
corresponding matrix element value identifies the gray level at
that point.

Continuous image projected Result of image sampling

4/3/2025 onto a sensor array (#/62) and quantization 26
 Digital Image Representation
• A digital image can be considered a matrix
whose row and column indices identify a
point in the image and the corresponding
matrix element value identifies the gray
level at that point. Pixel values in highlighted
region

4/3/2025 (#/62) 27
 Matrix Representation
183 160 94 153 194 163 132 165
183 153 116 176 187 166 130 169

179 168 171 182 179 170 131 167 
177 177 179 177 179 165 131 167 
178 178 179 176 182 164 130 171
 
179 180 180 179 183 169 132 169
179 179 180 182 183 170 129 173
 
180 179 181 179 181 170 130 169

Divide into
8x8 blocks
H=256

4/3/2025 W=256 (#/62) 28

 Grayscale image
• A grayscale or luminance image can be
considered to be one of the components of a
different parameterization.
• Advantage: It captures most of the “image
information”.
• Our emphasis in this class will be on general
processing. Hence we will mainly work with
grayscale images.

4/3/2025 (#/62) 29
 Gray level

• The amplitude of f at spatial coordinates (x, y) gives

the intensity (brightness) of the image at that point.
• we call the intensity of a monochrome image f at
coordinate (x,y) the gray level (l) of the image at that
point.
• Thus, l lies in the range [Lmin, Lmax]
• Lmin is positive and Lmax is finite.
• gray scale = [Lmin, Lmax]
• Common practice, shift the interval to [0, L-1]
• 0 = black , L-1 = white
4/3/2025 (#/62) 30
 Number of bits
• The number of
graylevels
typically is an
integer power of
N 2, L = 2k
• Number of bits
required to store
a digitized image
M b=MxNxk
4/3/2025 (#/62) 31
for i = 1 to 256
for j = 1 to 256
A(i, j) = j -1;
image(A);

image 256 × 256

4/3/2025 (#/62) 32
Circle with radius 80
for i = 1 to 256
for j = 1 to 256
dist = sqrt( ((i -128)ˆ2 +
(j -128)ˆ2))
if (dist < 80)
B(i, j) = 255;
else
B(i, j) = 0;
image(C);
4/3/2025 (#/62) 33
 Circle with radius 80
for i = 1 to 256
for j = 1 to 256
C(i; j) = A(i; j) * B(i; j)/255;
image(C);
Warning: Some processing will take an image A with
A(i; j) into a new matrix B which may not have integer
entries!
In these cases we must suitably scale and round the
elements of B in order to display it as an image.
4/3/2025 (#/62) 34
 Resolution
• Resolution (how much detail you can see)
depends on sampling and gray levels.
• the bigger the sampling rate (n) and the
gray scale (g), the better the
approximation of the digitized image from
the original.
• the more the quantization scale becomes,
the bigger the size of the digitized image.
4/3/2025 (#/62) 35
 Spatial Resolution

A 1024x1024, 8 bit image subsampled down to 32x32 pixels. Gray level 256
4/3/2025 (#/62) 36
 Pixel replica
a b c
d e f
(a) 1024x1024
(b) 512x512
(c) 256x256
(d) 128x128
(e) 64x64
(f) 32x32

if4/3/2025
the resolution is decreased
(#/62)
too much, the 37
checkerboard effect can occur.
 Zooming and Shrinking
• Zooming:
– Step 1: creates new pixel locations
– Step 2: assignment of graylevels to these
new locations.

4/3/2025 (#/62) 38
 Zooming
12 x 11  12 x 22
Nearest neighbor interpolation

Pixel replica a special case of NNI, possible only

when increase size is an integer multiple of
original size
11 22

12 12

4/3/2025 (#/62) 39
 Zooming
11 x 12  22 x 12

NNI suffers from checker board effect

22

12

4/3/2025 (#/62) 40
 Zooming
• Bilinear interpolation
f(x,y) = ax+by+cxy+d
here a,b,c,d are values corresponds to the
intensity of neighbor pixels

4/3/2025 (#/62) 41
 Zooming
a b c
zoomed to
1024x1024
from 128x128
64x64 32x32
nearest
neighbor
interpolation
d e f
by using
bilinear
interpolation

4/3/2025 (#/62) 42
 Shrinking
• Same as zooming, but reverse
• Row duplication row deletion
• Expand the grid and determine the
intensity
– Either by NNI, or
– Bilinear interpolation

4/3/2025 (#/62) 43
 False contouring
(a) Gray level = 16
(b) Gray level = 8
(c) Gray level = 4
(d) Gray level = 2
if the gray scale is not enough, the
smooth area will be affected.
False contouring can occur on the
smooth area which has fine gray
scales.
(a) (b)
(c) (d)

4/3/2025 (#/62) 44
 Nonuniform sampling
• for a fixed value of spatial resolution, the
appearance of the image can be improved
by using adaptive sampling rates.
• fine sampling  required in the
neighborhood of sharp gray-level
transitions.
• coarse sampling utilized in relatively
smooth regions.
4/3/2025 (#/62) 45
 Example

a b c: Low level details Mid level details High level details

4/3/2025 (#/62) 46
 Example
• an image with a face superimposed on a uniform
background.
• background little detailed information
coarse sampling is enough.
• face more detail fine sampling.
• if we can use adaptive sampling, the quality of
the image is improved.
• Moreover, we should care more around the
boundary of the object  sharp gray-level
transmission from object to background.
4/3/2025 (#/62) 47
 Nonuniform quantization
• unequally spaced levels in quantization
process influences on the decreasing the
number of gray level.
– use few gray levels in the neighborhood of
boundaries. Why ? eye is relatively poor to
estimate shades of gray near abrupt level
changes.
– use more gray levels on smooth area in order
to avoid the “false contouring”.
4/3/2025 (#/62) 48
 Basic Relationship between
pixels
• Neighbors of a pixel
• Connectivity
• Labeling of Connected Components
• Relations, Equivalences, and Transitive
Closure
• Distance Measures
• Arithmetic/Logic Operations

4/3/2025 (#/62) 49
 Neighbors of a pixel
• a pixel p at coordinate (x,y) has 4
– N4(p) : 4-neighbors of p 4 P 4
• (x+1, y), (x-1,y),(x,y+1), (x,y-1) 4
• Nd(p) : 4-diagonal neighbors of
p d d
• (x+1, y+1), (x+1,y-1),(x-1,y+1), (x- P
1,y-1)
d d
• N8(p) : 8-neighbors of p :
– a combination of N4 (p) and Nd (p) 8 8 8
8 P 8
4/3/2025 (#/62)
8 8 8 50
 Adjacency of pixels
• Let V be the set of gray-level values used to
defined adjacency
– 4-adjacency: two pixels p and q with values
from set V are 4-adjacent if q is in the set N4(p)
– 8-adjacency: two pixels p and q with values
from set V are 8-adjacent if q is in the set N8(p)
– m-adjacency: two pixels p and q with values
from set V are m-adjacent if
• q is in the set N4(p), or
• q is in the set Nd(p) and set N4(p)  N4(q) has no
pixels whose values are from V
4/3/2025 (#/62) 51
 Adjacency of pixels
•
V={1,2,3,4,5,6,7,8,9,10}
1 3 7
– {3} 4-adj.
11 6 13
– {1,3,7,9} 8-adj.
18 15 9
– {3} m-adj. (I)
– {9} m-adj. (II)
– {7} not m-adj. By {II}
– {11,13,18,15} not adj.

4/3/2025 (#/62) 52
 Adjacency (contd.)
• Two image subset S1 and S2 are adjacent if
some of pixels in S1 is adjacent to some of S2.
• A path from pixel p with coordinates (x,y) to pixel
q with coordinates (s,t) is a sequence of distinct
pixels with coordinates (x0,y0), (x1,y1),…(xn,yn)
where (x0,y0)= (x,y) , (xn,yn) = (s,t) and (xi,yi) is
adjacent to (xi-1,yi-1)
• n is the length of the path
• we can define 4-,8-, or m-paths depending on
type of adjacency specified.
4/3/2025 (#/62) 53
 Adjacency (contd.)

0 1 1 0 1 1 0 1 1
0 1 0 0 1 0 0 1 0
0 0 1 0 0 1 0 0 1
a b c
m-adjacent (8-path ambiguous) m-path no ambiguous

4/3/2025 (#/62) 54
 Definitions
• Let S be a subset of pixels and p and q
are said to be connected in S if there
exists a path between p and q which
completely lies in S.
• In S, a set of pixels which are connected in
S is called a connected component. If
number of component is 1, it is connected
set.

4/3/2025 (#/62) 55
 Definitions
• A connected set of pixels is called region
• Boundary (border or contour) of a region is
a set of pixels in the region that have one
or more neighbors that are not member of
the region.

4/3/2025 (#/62) 56
Labeling of Connected Components
(4-connected) [binary]
• Scan the image from left to right
• Let p denote the pixel at any step in the
scanning process.
• Let r denote the upper neighbor of p.
• Let t denote the left-hand neighbors of p.
• when we get to p, points r and t have
already been encountered and labeled if
they were 1’s.
r
4/3/2025 (#/62) t p 57
 Labeling of Connected
Components
• if the value of p = 0, move on.
• if the value of p = 1, examine r and t.
– if they are both 0, assign a new label to p.
– if only one of them is 1, assign its label to p.
– if they are both 1
• if they have the same label, assign that label to p.
• if not, assign one of the labels to p and make a note that the two
labels are equivalent. (r and t are connected through p).
• at the end of the scan, all points with value 1 have
been labeled.
• do a second scan, assign a new label for each
equivalent labels.
4/3/2025 (#/62) 58
 Component labeling (4)
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 1 0 0 1 1 1 0 1 1 0 0 2 2 2
0 1 1 0 0 0 1 1 0 1 1 0 0 0 2 2
0 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0
0 0 0 1 1 0 0 0 0 0 0 3 3 0 0 0
0 0 0 1 1 0 0 0 0 0 0 3 3 0 0 0
0 0 0 1 1 0 0 0 0 0 0 3 3 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1 1 1 0 1 0 0 0 1 1 1 0 1 0 0 0
0 0 1 1 1 0 0 0 0 0 1 1 1 0 0 0
0 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 1 1 0 1 1 0 0 2 2 2 0 3 3 0 0
1 1 1 0 0 1 1 0 2 2 2 0 0 3 3 0
0 0 0 0 0 1 1 1 0 0 0 0 0 3 3 3
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 Labeling of Connected
Components (8-connected)
• do the same way but examine also the upper
diagonal neighbors of p.
• if p is 0, move on.
• if p is 1
– if all four neighbors are 0, assign a new label to p.
– if only one of the neighbors is 1, assign its label to p.
– if two or more neighbors are 1, assign one of the label
to p and make a note of equivalent classes.
• do the second round and introduce a unique
label to each equivalent class.

4/3/2025 (#/62) 60
 Component labeling (8)
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 1 1 0 0 1 1 1 0 1 1 0 0 2 2 2
0 1 1 0 0 0 1 1 0 1 1 0 0 0 2 2
0 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0
0 0 0 1 1 0 0 0 0 0 0 1 1 0 0 0
0 0 0 1 1 0 0 0 0 0 0 1 1 0 0 0
0 0 0 1 1 0 0 0 0 0 0 1 1 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0

1 1 1 0 1 0 0 0 1 1 1 0 1 0 0 0
0 0 1 1 1 0 0 0 0 0 1 1 1 0 0 0
0 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1 1 1 0 1 1 0 0 2 2 2 0 3 3 0 0
1 1 1 0 0 1 1 0 2 2 2 0 0 3 3 0
0 0 0 0 0 1 1 1 0 0 0 0 0 3 3 3
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
 Distances between pixels
• For pixels p, q and z say, (x,y), (s,t) and
(v,w) be the locations. The distance
function is
– D(p,q) >=0 ( =0 if p=q)
– D(p,q) = D(q,p)
– D(p,z)<= D(p,q) + D(q,z)
• Euclidean distance De (p,q) = ( x  s) 2  ( y  t ) 2

4/3/2025 (#/62) 62
 Distances
• D4 distances (city block distance)
|x-s| + |y-t|

2
2 1 2
2 1 0 1 2
2 1 2
2

4/3/2025 (#/62) 63
 Distances
• D8 distances (chessboard distance)
max(|x-s|,|y-t|)

2 2 2 2 2
2 1 1 1 2
2 1 0 1 2
2 1 1 1 2
2 2 2 2 2

4/3/2025 (#/62) 64