GIS Architectures, lecture 7

1
GIS Architectures, lecture 7

• Digitization
• digitization is necessary if data is not yet in
  digital format
• nowadays data is often already in digital format

2
Digitization

• Scanning
• Raster to Vector conversion
• Optical Character Recognition (OCR)
• On-screen digitization
• Digitization with a tablet or table

3
(Image) scanners

• A device which analyzes a physical image or an
  object and converts it into a digital image
• CCD sensor (photoreactive surface capacitors)
• hand-held scanners (not used anymore)
• flatbed/desktop scanners
• PMT sensor (photomultiplier tube)
• drum scanner
• CMOS sensor (photoreactive surface
  semiconductors)
• not used in scanners since low quality

4
CCD, Charge-coupled Device
photon
photoreactive surface pixel
capacitor
light (photons) release electrons from the
photoreactive surface, which accumulate in the
capacitor creating a charge which is relative to
the amount of light that has fallen on the
surface
5
Color sensing with CCDs
Bayer mask
respective CCD pixels measure the amount of red,
green or blue light falling on the surfaces
there are two green pixels since human eye (cone
cells) are more sensitive to green
another (more expensive) possibility is to use
three CCDs and a prism
6
CCD array grid
array of pixels capacitors
Electronics
grid of pixels
the electronics measure the charge, empty the
capacitors, digitise the information and store it
Electronics
7
PMT

• Extremely sensitive detectors of light

8
Specifications for scanners 1.

• Size of original (width, A4, ..), type of
  original
• B/W, grayscale or color
• color depth (bits)
• Speed (cm/s)
• Resolution (SPI, samples per inch)
• look at optical resolution since software
  interpolated values are meaningless

9
Specifications for scanners 2.

• Connectivity (Parallel port (slow), SCSI
  (faster), USB 1 (relatively slow), USB 2
  (fast)... Driver software)
• Included software
• Output formats
• Price

10
Drum scanners

• The image is wet-mounted (in oil) on a cylinder
• Full-spectrum light is beamed through the image
  or reflected from the image
• Cylider is rotated and PMT records one pixel at a
  time
• High-quality instruments
• expensive 5000 - 40 000
• require monitoring and calibration
• optical resolution is 8000-14000 spi

11
CCD scanners

• Mass-production scanners
• Sensor is typically a line (or three lines for
  color scanners) of several thousand CCDs
• Light source is cold cathode (a neon light
  variant) lamp or LED (light emitting diode)
• Optical resolution of a CCD scanner is
  1200..1600-3200..5400 spi
• Typical cost 50 - 2000
• High-end CCD scanners may be as expensive as drum
  scanners (and of comparable quality)

12
Comparison
Drum
CCD
http//www.marginalsoftware.com/Scanner/downtown.h
tm
13
Scanner output

• typically uncompressed RGB image transferred to
  host computer
• can be very large (thus requires a fast
  connection)

14
TWAIN Technology Without An Interesting Name

• An API for Windows and Mac
• A program running in the host computer makes
  requests (get image, delete image, ...) to the
  peripheral
• Contains an architecturel blunder Does not
  separate user interface from the device driver
• makes it difficult to provide network access
• an application using a TWAIN driver has to use
  the manufacturers GUI also
• especially in digital cameras the images cant
  be retrieved as files from the source

15
SANE (Scanner Access Now Easy)

• An API that provides standardized access to any
  raster image scanner hardware
• SANE is open and free technology
• saned allows network access to the scanner which
  is connected to a computer

16
Raster to Vector conversion

• Difficult (to program) and error-prone process

Convert to internal raster representation
Rectification
Remove errors (trash) from the image
Smooth the lines in the image
Make continuous lines continuous
Thin the lines
Find the lines (straight or curves?)
Manual editing
17
Rectification

• On-screen digitization of known points
• xf(x,y)
• yf(x,y)
• decide on the target raster
• go through the target raster pixel by pixel and
  look from the original image what to put in them

18
Error removal

• Remove areas which are smaller than some
  threshold
• dashed lines?
• Image manipulation methods
• grayscale thresholding,

19
Thinning

• Jang, B-K., Chin, R.T. 1990. Analysis of Thinning
  Algorithms Using Mathematical Morphology. IEEE
  Trans. Pattern Analysis and Machine Intelligence.
  12(6). 541-551.
• used by, e.g., GRASS
• The thinning algorithm defines a set of
  structuring templates and applies them in several
  passes until there are no matches or until the
  maxiterations is reached. Trimming means certain
  structuring templates are applied to kill
  emerging short limbs which appear because of
  noise in the raster.

20
Templates (in the thinning algorithm)

• A 3 by 3 matrix (pixel neighborhood) which is
  applied on the source raster.
• Template has values
• 0 and 1 these must match the pixels in the
  source raster
• -1 dont care about this pixel
• if the template matches, the the target raster
  will have a specified new value, otherwise the
  value from the source raster is used

21
OCR

• Optical character recognition
• Scanned text ? text
• Complications
• different fonts
• alignment
• Often based on trained neural networks

22
Digitization tablet or table

• Specifications
• Size of the work area (44x60 112cmx152cm
  -gtsmaller)
• Accuracy (0,005 0,13mm -gt lower)
• Connectivity (physical, logical)
• Compatible software
• Price (..1500-2500.. for large digitizers)

23
Digitization

• A thriving industry in 3rd world countries (cheap
  semi-skilled labor)