Digital Video

30 frames per second.

DV Cameras introduced in 1995.

Commonly using 1/4” - 1/3” CCDs with 300k-1M pixels.

Single CCD vs. 3 CCDs (better quality)

DV obtained by:
ƒ Sampling an analog video signal V(t)
ƒ Sampling the 3-D space-time intensity distribution I(x,y,t)

Video Sampling

Progressive scanning: One full frame every

1/30th of a second.

Interlaced scanning: two separate fields every

1/60th of a second.
(P:1 interlacing)

1
 Progressive scanning

Frame tn:

Interlaced Scanning
Frame tn: Frame tn+1:

2
 Motion effects
(Sawtooth type effects)

Interlaced video:

De-interlaced:

Motion effects
(Sawtooth type effects)

3
 Interlacing effects

Interlacing effects

Line flicker
(due to variations in brightness
between successive fields)

http://www.progressivescan.co.uk/images/progressive03.gif

4
 Basic Image Transformations

Image Processing: Input: image; Output: image.

Image Analysis: Input: image; Output: information.

Image Operations:
Point, algebraic, geometric operations

Image Histogram

Image f(x,y) x:[0,N-1] y:[0,M-1] (N rows, M columns)

Quantization: k levels [0,k-1] (k gray levels)

Image histogram Hf : a plot of the frequency of occurrence of every gray

level n.

Ki=0,k-1
Hf(ki)=J
J=: # pixels with gray value ki

5
 Point Operations
Point operation is a function h applied identically to every pixel in an input image
f(x,y) to create anew modified image g(x,y):

g(x,y)=h[f(x,y)]

Linear point operations:

g(x,y)=Pf(x,y)+S

P>1 …… histogram expansion

P=1 …… histogram shift (e.g. creating a brighter or darker image)
P<1 …… histogram shrinkage

Point Operations (cont.)

Example outcomes:
Ξ Brighter image
Ξ Darker image
Ξ Negative image
Ξ Contrast stretch

6
 Point Operations (cont.)
Example outcomes:
Ξ Brighter image P=1, S>0
Ξ Darker image P=1, S<0
Ξ Negative image P=-1, S=k-1
Ξ Full Scale Histogram Stretch P=(k-1/max-min), S=min*(k-1/max-min)

Non-Linear Point Operations.

Examples:
Ξ Logarithmic operations
g(x,y) = FSHS {log[1+f(x,y)]}
to improve the visibility of dim objects.
Ξ Histogram equalization

Arithmetic Operations
Operations defined over multiple images.

Examples:
ƒ Pointwise image addition:
g=f1+f2+…+fn=Ζf
ƒ Pointwise image differencing:
g=f2-f1
ƒ Image averaging for noise reduction: g=(1/n)∑fi

7
 Geometric Operations
Modify the location and relationships of pixels within an image, without
modifying their gray values.

g(x,y)=f(x’,y’)=f[h(x,y)]

Gray value assignment:

α nearest neighbor interpolation
α bilinear interpolation

Example geometric operations:

α Translation
α Rotation
α Zoom

Image Analysis Tools

Key concepts:
– Filtering
– Convolution
– Edges
– Fourier

8
 Filtering & Convolution
A mask is applied onto the image. The basic approach is to sum
products between mask coefficients and corresponding pixel
intensities at a specific location in the image:

R = w1z1 + w 2 z2 + ...+ w 9 z9

Filtering (cont.)
Averaging filters blur an image, e.g.
1 1 1
(low pass filtering)
1 1 1
0
1 1 1

Sharpening filters:
e.g.
(high pass filtering) -1 -1 1
-1 8 1
-1 -1 1

9
 Filtering (cont.)
Derivative filters emphasize edges
as maxima of first derivatives or -1 0 1
minima of second derivatives, e.g.
-1 0 1
-1 0 1

Filtering (cont.)
Second derivative filters e.g. Laplacian:

∂2 f ∂2 f
∇2 f = +
∂x 2 ∂y 2

∇ 2 f = 4 z 5 − ( z 2 + z 4 + z 6 + z8 )

10
 Filtering (cont.)
LoG filter:

 x2 + y2 
G(x, y) = exp− 2 
 2σ 

 r2 − σ 2   r2 
LoG : ∇ h = 
2
 exp− 2 
 σ   2σ 
4

Fourier Transform
Image f(x,y) has an equivalent in the frequency domain F(u,v).
Operations in the spatial domain have an equivalent in the
frequency domain.

11
 Fourier Transform
Filters

Ideal low-pass:
1 if D(u , v) ≤ D0
H (u , v) = 
0 if D(u , v) > D0

D(u, v) = (u 2 + v 2 )1/ 2

12