1. Digital image basics Human vision perception system Image formation
Human vision property and model
Image acquisition
Image transform
Image quality
Connected components
Image sensing
Image formats

2. 1. Human vision

3. 1. Human vision (cont’d)
Two types of receptors
-- Cones (fovea): sensitive to brightness and color
- 7M
- Cone-vision (photopic, bright-light vision)
-- Rod (cell): sensitive to low-level illumination
- 100M
- Rod-vision (scopotic, dim-light vision)

4. 2. Image perception and formation

5. 3. Vision property

6. 3. Vision property (cont’d)
Brightness adaptation
-- There are a range of intensity levels that human eye can adapt
- photopic: 10^(-3) (mL) – 10^(3) (mL)
- scopotic: 10^(-3) (mL) – 10^(-1) (mL)
-- Human eyes have brightness adaptation level,
they cannot adapt the whole range
simultaneously

7. 3. Vision property (cont’d)
Brightness discrimination
-- The ability to discriminate different intensity level
- Weber ratio: just noticeable difference of intensity versus the background intensity
-- The intensity defined in the digital image is not the real intensity. It is a contrast scale (e.g., gray scale)

8. 3. Vision property (cont’d)
Contrast
-- Absolute contrast
C = Bmax / Bmin
where Bmax is the maximum brightness intensity
Bmin is the minimum brightness intensity
-- Relative contrast
Cr = (B – B0) / B0
B is the brightness of object; B0 is the background brightness
-- Mach Band: over-shooting effect

10. 3. Vision property (cont’d)
Spatial discrimination (SD)
-- minimum view angle which can discriminate two points on the object to be viewed
d/(2 * Pi * L) = theta / 360 d L
theta
eye

11. 3. Vision property (cont’d)
Spatial discrimination (SD)
-- low illumination (SD decreases)
-- low contrast (SD decreases)
-- too high illumination (SD does not increase too much)
-- SD of color is weaker than SD of brightness
-- projection on fovea (SD increases)

12. 3. Vision property (cont’d)
Human vision model
-- g(x, y) = T [ f(x, y)]
-- T: transform input optical scene to output image
- linear or non-linear transform
- H(u,v) low pass filter (e.g., limited discrimination, linear)
- log response to the brightness (e.g., non-linear)
- time-delay effect (e.g., “image-remain” effect)
Input image f(x,y)
Optical system H(u,v) ~ h(x,y)
Output image g(x,y)

13. 4. Image acquisition
Wavelength
-- electromagnetic spectrum

14. 4. Image acquisition (cont’d)
Principle of imaging sensor
-- transform illumination energy into digital image
-- output voltage waveform is proportional to light
-- e.g., single sensor, group sensors (one-strip, CT/MRI), group sensors (2D array CCD)

15. 4. Image acquisition (cont’d)
Image digitizing
-- Sampling: digitizing the coordinate values (spatially)
- Nyquest rate: 2*F(max)
- limited by the number of sensors
- spatial sampling: uniform and non-uniform
(e.g., fovea-based, fish-eye based)
-- Quantization: digitizing the amplitude values
- uniform
- non-uniform (based on image characteristics)

16. 4. Image acquisition (cont’d)
Image digitizing
-- f(x, y) is the gray level at pixel location (x, y)
-- Gray level is not real illumination intensity (it is an
index of the gray scale)
-- f(x, y) is in the range of [0, 255] for 8-bit image
-- the image with size of M*N and k bits per pixel,
has the total bits: M*N*k

17. 4. Image acquisition (cont’d)
Spatial resolution
-- number of pixels with respect to the image size
-- line pair: smallest discernible detail per unit
distance in an image
- e.g., 100 lp/mm.

18. 4. Image acquisition (cont’d)
Relationship between spatial resolution N and gray level resolution K
-- N  and K   quality 
-- N  and K   contrast 
-- N (detail)   K (number of gray level) can be 
(e.g., half-tone image)

19. 4. Image acquisition (cont’d)
Aliasing problem
-- JigJag or staircase effect.
-- occurs in image acquisition (e.g., image processing)
-- occurs in display (e.g., computer graphics)
-- Reasons:
The sampling or displaying resolution is lower than the
minimum rate 2*F(max), which is the Nyquest rate.
-- Possible solution:
- Smooth image before sampling to reduce the F(max)
- side-effect: image blurred, quality 

20. 5. Image transform Size change -- Zoom-in Shape change -- Zoom-out
-- pixel replication
-- pixel interpolation
-- super-resolution
Shape change
-- geometric transformation

21. 6. Image quality Subjective -- Rating (e.g., R=1, 2,…, 5)
where N is the number of evaluators; Ji R
-- application in image enhancement, restoration, compression, etc.

22. 6. Image quality Objective -- Mean square error -- dB value: -10Log(E)
-- f(x,y) is the image to be evaluated.
f^(x,y) is the reference image to be compared with.
-- application in image coding, etc.

23. 7. Connected components Relationship of pixels
-- Four neighbors of pixel P
- N4(P) (strong neighbors)
- ND(P) (weak neighbors)
-- Eight neighbors of pixel P
- N8(P) = N4(P) + ND(P) P P P
Strong weak neighbor

24. 7. Connected components (cont’d)
Adjacency
-- 4-adjacency
-- 8-adjacency
-- m-adjacency (mixed-adjacency) q P P q P q
4-connected pq is not m-connected connected
m-adjacent: if q is N4(p), or
q is Nd(p) and N4(p) N4(q) = 

25. 7. Connected components (cont’d)
Path
-- If p and q is connected, there is a path between p and q.
-- m path: the path between p and q based on m-connected pixels.
-- closed path: starting p and ending q are connected P q P q

26. 7. Connected components (cont’d)
-- set of pixels which are connected
-- The set is also called connected set
Concept
-- R is a region if R is a connected set
-- boundary of R is “closed path”
-- edge: gray-level discontinuity at a point
- link edge points  edge segment

27. 7. Connected components (cont’d)
Distance
-- D(p, q) is defined as the distance between p and q.
D(p, q) >=0
D(p, q) = D(q, p)
D(p, q) <= D(p,z) + D(q,z)
-- Euclidean distance (disk shape)
De(p,q) = sqrt[(xp – xq)^(2) + (yp – yq)^(2)]

28. 7. Connected components (cont’d)
Distance
-- D4 distance (city-block distance) (diamond shape)
D4(p,q) = |(xp – xq)|+ |(yp – yq)| 2

29. 7. Connected components (cont’d)
Distance
-- D8 distance (chessboard distance) (square shape)
D8(p,q) = max(|(xp – xq)|, |(yp – yq)|)

30. 7. Connected components (cont’d)
Distance
-- Dm distance (shortest m-path between two points)
1 - 1 | 1
Dm = 4

31. 8. Pixel operation Point-wise operation -- M*N image bound matrix t r
(r,t): coordinates of upper-left component; each component is either defined (which is represented by a certain intensity value), or undefined (which is represented by “*”).

32. 8. Pixel operation (Cont’d)
Arithmetic operation
(1) ADD[f, g](I,j)
= f(I,j) + g(I,j) IF f(I,j)   and g(I,j)   (C1)
=  otherwise
(2) Mult[f,g](I,j)
= f(I,j) • g(I,j) IF C1
(3) SCALAR[t; f](I,j)
= t • f(I,j) IF f(I,j)  
=  otherwise

33. 8. Pixel operation (Cont’d)
Arithmetic operation
(4) Max[f,g](I,j)
= max[f(I,j), g(I,j)] IF C1
=  otherwise
(5) Min[f,g](I,j)
= min[f(I,j), g(I,j)] IF C1
(6) Sub[f](I,j)
= -f(I,j) IF f(I,j)  
(6) SCALAR[t; f](I,j)
= t • f(I,j) IF C1
=  otherwise

34. 8. Pixel operation (Cont’d)
Arithmetic operation
(7) EXTEND[f,g](I,j)
= f(I,j) IF f(I,j)  
= g(I,j) otherwise
(8) EXTADD[f,g](I,j)
= ADD[f,g](I,j) IF C1
= f(I,j) IF f(I,j)   and g(I,j) = 
= g(I,j) IF g(I,j)   and f(I,j) = 
= * both g and f on undefined domain

35. 8. Pixel operation (Cont’d)
Arithmetic operation
(9) THRESH[f,t](I,j)
= IF f(I,j)  t
= IF f(I,j) < t
=  IF f(I,j) = 
(10) TRUNC[f,t](I,j)
= f(I,j) IF f(I,j)  t
TRUNC[f,g](I,j) = Mult[f, THRESH(f, t)]

36. 8. Pixel operation (Cont’d)
Arithmetic operation
(11) EQUAL[f,t](I,j)
= IF f(I,j) = t
= otherwise
= * on the undefined domain
(12) similar definition for
GREATER[f,t](I,j)
BETWEEN[f, t1, t2](I,j)
(13) operation with masking:
AND, OR, NOT.

37. 8. Pixel operation (Cont’d)
Arithmetic operation
(14) PIXSUM(f) is the summation of all pixels on the
defined domain
(15) DOT(f,g) = SUM[f(I,j)  g(I,j)] on the common domain
(16) Norm(f) = [SUM[f(I,j)^2]]^(1/2)
Norm(f) = (DOT(f,f))^(1/2)

38. 8. Pixel operation (Cont’d)
Arithmetic operation
(17) REST[f,g](I,j)
= f(I,j) IF g(I,j)  
=  IF g(I,j) = 
(18) Note:
Linear operation: H(af + bg) = aH(f) + bH(g)
otherwise: non-linear operation (e.g., |f-g| operation)
H: operator
f, g: images
a, b: scale values

39. Image Sensing Single Image Sensor Line Sensor (Sensor strip)
Array Sensor

40. Image Sensing Linear motion Rotation
Sensing Ring for CT (x-ray) to create cross-sectional images

41. Image Format
TIF (LZW – lossless coding)
GIF
JPEG
BMP

42. Image Format TIF (LZW – lossless coding) Tagged image file format
Image head:
field = tags + values
image size
compression
color depth
location of data
bits per sample
…….

43. Image Format
JPEG
8*8 blocks  DCT  Coefficient quantization  Huffman coding  zig-zag run-length coding

44. Demo

45. Image Format BMP PBM - portable bitmap file format (binary)
PGM – portable greymap (grey scale)
PPM – portable pixmap (color)