Digital Image Processing (Digital bildbehandling grundkurs) Digital ...

p. 1

Digital Image Processing (Digital bildbehandling grundkurs)

Digital Image Processing Lecture 1



Properties of the digital image The digital image sensor and image acquisition Sampling and quantization Spatial and Intensity resolution The electromagnetic spectrum, imaging and examples of images from different sources Gray scale and color images Color maps The eye and two visual phenomena Application examples



Gonzales & Woods:

    

Maria Magnusson, Computer Vision Laboratory, CVL Department of Electrical Engineering (EE) Linköping University

  

Maria Magnusson, Avdelningen för Datorseende Institutionen för Systemteknik (ISY) Tekniska Högskolan Linköpings Universitet

A continuous image is 2D function 

1D: f(t) is a function f that depends on the time t. 2D: f(x,y) is a function f that depends on the spatial (room) coordinates x and y.



Ex)



f  x, y   sin x  y   1 f  x, y   0  black

f  x, y   2  white

  

p. 2

Chapter 1: pp. 1-34 (Paperback 2013: Add 22 to the page numbers) Chapter 2: pp. 35-64 (Paperback 2013: Add 22 to the page numbers) Extra

Maria Magnusson, Computer Vision Lab., Dept. of Electrical Engineering, Linköping University

p. 3

An example of a digital image pixel = picture element

zoom 

Original image size: 720x240 pixels

p. 4

p. 5

The digital image sensor

p. 6

The digital image acquisition process

Single sensor element:

Line sensor:

Array sensor:

typically a camera

Fig. 2.12

p. 7

Continuous Image

Line from image

After sampling

After quantization Fig. 2.16

The digital image sensor

For a color image, there are 3 different types of sensor elements: red, green and blue

p. 8

Senseorelement

Image sampling and quantization

Fig. 2.15

A continuous image projected onto the sensor array

Result after sampling and quantization

Fig. 2.17

p. 9

Representation of a digital image

p. 10

Saturation and noise Saturation is the highest value beyond which all intensity levels are clipped.

3D surface plot

Noise appears often as a grainy texture. Intensity image

Numerical matrix

Fig. 2.18

p. 11

Spatial resolution 



Spatial resolution is a measure of the smallest discernible detail in the image. Two common measures: lp/mm, dpi

5 discernible line pairs (lp) per mm, 5 lp/mm

   

Newspaper: 75 dpi Magazine: 133 dpi Glossy brochures: 175 dpi Gonzalez & Woods:2400 dpi

6 dots per inch, 6 dpi

1 inch 0.1 mm 1 mm

Fig. 2.19

Spatial resolution

p. 12

1250 dpi 3692 x 2812 pixels

300 dpi 886 x 675 pixels



A newspaper image 6 pixels

Fig. 4.22

Fig. 2.20

p. 13

Intensity resolution 28 intensity levels

Fig. 2.21

24

23

212 intensity levels

27

p. 14

The same CT image, shown in 2 different contrast windows

26

22

25

21

p. 15

The electromagnetic spectrum

Intensity resolution  





 

26=64

We can only resolve (or intensity levels on an ordinary computer screen. Based on hardware considerations, gray scale images are usually stored with 8 bits per pixels, i.e. 28=256 intensity levels. Some images are stored with more than 8 bits per pixels. CT images, for example, are stored with 12 bits per pixels, i.e. 212=4096 intensity levels. During calculation and image processing, pixels can preferably be stored with more than 8 bits or floating point numbers. Special hardware utilized for fast calculation sometimes store a pixel with less than 8 bits. Color images are usually stored with 3x8 bits per pixels, 28 red intensity levels, 28 green intensity levels and 28 blue intensity levels giving 224 = 16 million different colors.

micro waves

p. 16

radio waves

27=128?)

Visible light





Common Cameras Se also Fig. 2.10  blue: around 0.445mm , green: around 0.535mm, red: around 0.575mm  IR: 8-14 mm Thermal bands in NASA’s LANDSAT satellite  1. Visible blue, 0.45-0.52mm. Maximum water penetration.  2. Visible green, 0.52-0.60mm. Good for measuring plant vigor.  3. Visible red, 0.63-0.69mm. Vegetation discrimination.  4. Near infrared, 0.76-0.90mm. Biomass and shoreline mapping.  5. Middle infrared, 1.55-1.75mm. Moisture content of soil & vegetation.  6. Thermal infrared, 10.4-12.5mm. Soil moisture; thermal mapping.  7. Middle infrared 2.08-2.35mm. Mineral mapping.

p. 17

Images from NASA’s LANDSAT satellite Fig. 1.10

Imaging – the creation of images Examples of different sources 

Electromagnetic radiation   

6. Thermal infrared

4. Near infrared

     

1. Visible blue

2. Visible green



3. Visible red

p. 19

Image Reconstruction:

To compute an image from the in-signal. Ex: Computed Tomography (CT). X-ray Source

Curved detector with projection data

Patient Rotation

Reconstruction algorithm (mathematics) Reconstructed image

p. 18

Transmitted gamma-rays, fig. 1.6a Transmitted and reconstructed gamma-rays, Single Photon Emission Tomography, SPECT Emitted and reconstructed gamma-rays, Positron Emission Tomography, PET, fig. 1.6b, slide 28 Transmitted X-rays, fig 1.7a,b Transmitted and reconstructed X-rays, Computed Tomography, fig. 1.7c, slide 14, 19 Excitation rays from ultraviolet light, fluorescence microscopy, fig. 1.8a,b Reflected rays from visible light, slide 4, 10, 12, 23 Reflected infrared rays, slide 17, Emitted radio waves + magnetic fields and reconstruction => magnetic resonance imaging (MRI), fig 1.17

Ultrasound 

Transmitted and reflected ultrasound, fig 1.20

Examples of different images from different sources

p. 20

Fig. 1.6a

Fig. 1.6b

Transmitted gammarays

PET

Examples of different images from different sources Fig. 1.7ab

p. 21

Examples of different images from different sources

p. 22

Fig. 1.8ab

Transmitted X-rays

Fluorescence microscopy

Fig. 1.17 MRI

Fig. 1.20ab Transmitted and reflected ultrasound

Gray scale image compared to true color image Matlab : Colim  zeros(450, 600,3); Colim(:, :,1)  R; Colim(:, ; ,2)  G; Colim(:, :,3)  B; True color image: Colim

p. 23

Landsat’s images combined to color images

p. 24

Fig. 6.27

Fig. 2.38 B G R Gray scale image: Grayim

Grayim  0.2989  R  0.5870  G  0.1140  B; Grayim  rgb2gray(Colim);

3, 2, 1 = R,G,B combined to color image

4, 2, 1 = R,G,B combined to pseudo-color image

p. 25

Color maps 

Sometimes, the pixels are quantified to the interval [0,255]. These values are transformed through a color map in the computer to  





Gray scale values, i.e. 0->black and 255->white or Arbitrary colors (pseudo colors)

Sometimes, the pixels are floating point values. These values are transformed to the interval [0,255] and further through a color map in the computer. A true color image has 3 values per pixel. These values are transformed individually through 3 color maps in the computer and further to the red, green and blue channel of the screen.

True Color map Pixel value [fr(x,y),fg(x,y),fb(x,y)]

Normal gray scale color map

Linear transformation

0: 1: 2:

255:

255

To D/A-converter and further to the red channel of the screen.


0: 1: 2:

255:

G 0 1 2

255

To D/A-converter and further to the green channel of the screen.


256 colors

Pixel value f(x,y) 0: 1: 2:


A D/A-converter converts a digital value to an analog value, an electrical voltage.

255:

R 0 1 2

G 0 1 2

B 0 1 2

255

255

255

To D/A-converter and further to the screen.

p. 27

Over 16 million colors

p. 28

256 colors

Pseudo-color map Pixel value f(x,y)

R 0 1 2

p. 26

0: 1: 2:

255:

B 0 1 2

255

To D/A-converter and further to the blue channel of the screen.


R

G

B

?

?

?

0: 1: 2:

255: To D/A-converter and further to the screen.

Ex) A PET-image shows where it is activity in the brain. High activity is shown red and low activity is shown blue.

p. 29

Additive color mixing.

CRT LCD plasma

Basically, the eye works rather much like a camera…  Reflected light from the palm tree

yellow  red  green

Lens

p. 30

The retina is the sensor

cyan  green  blue magenta  red  blue Lab 5 Lecture 11: More on color

Focal length 

But the brain performs some more processing that gives rise to visual phenomena…

Fig. 6.4

Visual phenomenon 1: The Mach band effect

p. 31

Fig. 2.3

Visual phenomenon 2: Simultaneous contrast

p. 32

All the inner squares have the same intensity, but they appear progressively darker as the background becomes brighter, which are seen when the red frames are removed.

Gray scale Image:

Actual intensity:

Perceived intensity: Fig. 2.7

Fig. 2.8

p. 33

Application examples: Automated visual inspection of manufactured goods

p. 34

Some more application examples Paper currency

Finger print Controller board

Pill container

Bottles filling level

Automated license plate reading

Air pockets in plastic

Fig. 1.15

Fig. 1.14

The book and the course Multidimensional Signal Analysis TSBB06 included partly included not included Image sensors TSBB09

Image and Audio Coding TSBK02

p. 35

Fig. 1.23





Reconstruction

p. 36

Question for G  f   g t    g t  e  j 2p f t dt  next lecture: What is negative frequency in the Fourier transform?  



1. sin(−2pu) is an example of a signal with negative frequency. 2.− sin(2pu) is an example of a signal with negative frequency. 3. The frequencies in the Fourier transform have no relation to real frequencies. 4. …