Geography 38:376 Intro GIS

1
Geography 38376Intro GIS

• Topic 4
• Raster Vector Conversion
• Parts of Chapters 5 6 Chang
• (Chapter 5 DeMers)

2
Types of Conversion

• Rasterisation
• Vector to raster conversion
• Vectorisation
• Raster to vector conversion

3
(No Transcript)
4
Why Convert?

• Often necessary to convert between the two
  primary spatial data structures in order to
• Rasterise vector data when performing analysis
  that requires statistical or algebraic operations
• Vectorise classified RS imagery to extract
  thematic information
• (e.g. transportation networks, hydrologic
  features, landcover classes)
• Vectorise scanned maps (ArcScan)
• When importing existing data sources

5
Why Integrate?

• Raster data frequently used as image background
  for vector data display
• Raster images can also be linked to vector
  features hotlinks
• Vector data used to subset and classify raster
  satellite imagery
• Raster-vector data still seldom used in analysis
  together

6
Rasterisation

• Typically, all that is required is that the user
• specifies the output grid cell size
• and attribute to be used as a cell value
• Selection of grid cell size v. important
• Goal is to maintain feature information
• For polygon rasterisation a good rule of thumb is
  to select cell size ½ min. dimension of smallest
  feature that should be preserved

7
Rasterisation

• Any rasterisation algorithm identifies a set of
  grid cells that coincides with the point, line or
  polygon features being rasterised
• There are different ways of doing this
• Inevitably results in lose of info and
  distortion
• Absolute location and extent (size) of a point
• Position, length, width of a line
• Extent, shape, area of a polygon
• Smaller cell size, less severe errors, but larger
  file size

8
Ex. Rasterising a Straight Line

• Locate end pts of line in raster co-ordinate
  system
• Determine number of pixels required to complete
  line
• Given number of steps, determine x/y increment
• Starting at first pixel, add x/y increment to
  find next pixel, round off, set pixel, find next
  pixel

9
Bresenhams Line Algorithm

• Assumptions
• Once start and end pixels are identified
• . . . and number of pixels are determined
• Next pixel is always choice between one of two
  pixels

10

• If slope lt 45? - pixel to right or diagonally up
  to right
• If slope gt 45? - pixel above or diagonally up to
  right
• Choice is made based on decision variable, D
• D is based on slope (error term) and incremented
  each time
• Much faster than previous algorithm

11
Polygon Rasterisation

• First, how do algorithms handle boundary pixels?
• Central Point
• Includes a pixel if centre is inside polygon
  boundary
• Dominant Unit
• Includes a pixels if more than half its area is
  inside polygon
• Ranked List
• Decision is based on a user defined ranking of
  values
• At a given grid cell size, error is roughly
  equivalent
• Error increases with cell size boundary
  complexity

12
Central Point
Dominant Unit
13
Scan Line Coherence

• Most common polygon rasterisation algorithm

Fill grid cells b/w boundary cells
14
Vectorisation

• Most commonly performed when
• Extracting features from scanned maps or photos
• Extracting features from classified RS imagery
• High-res flatbed scanners are relatively
  inexpensive
• Large format scanners more common, but still
  costly
• Classified RS data is likely the second largest
  source of geomatics data

15
(No Transcript)
16
Vectorisation

• When scanning a map or photograph consider
• Colour depth
• Significantly affects file size
• Enables discrimination b/w features
• Use lowest possible
• Resolution
• Significantly affects file size
• Determines level of detail
• Use highest you can afford

17
Automated Vectorisation

• Once a raster image has been acquired
• Obtain bi-level or boolean image
• In the case of scanned maps or photos this may
  require some image pre-processing
• Noise reduction
• Edge enhancement
• Thresholding

18
Vectorisation

• If scanned as a grey scale image, it can be
  converted by decreasing the colour depth to 1-bit
  (B/W)
• Often results in gaps or clutter/noise depending
  on the quality and number of grey levels in the
  original
• Thresholding enables the user to select a
  cut-off value that distinguishes foreground
  from background pixels

19
(No Transcript)
20
Vectorisation

• In the case of classified RS imagery
• A boolean image is obtained by reclassifying data
• Desired class value is set to 1, everything else
  zero

21
Vectorisation

• Editing/pre-processing boolean image
• Filling may be required to remove gaps
• Manual editing to delete noise
• Line thinning algorithms
• Identify skeletal pixels
• Reduce point, line, and polygons to a single
  pixel width
• Balance erosion/deflation reduces error in final
  position

22
(No Transcript)
23
Vectorisation

• Vectorisation of boolean image
• Vector representations determined by establishing
  a node or vertex at the centre of each grid cell
• Node detection algorithms use roving windows to
  identify the beginning, end, and intersection of
  lines

24
Vectorisation

• Post-processing vector coverage
• Cleaning the coverage (manual and/or
  automatically)
• Dangling nodes
• Pseudo-nodes
• Line thinning (weed tolerances)
• Line smoothing (splinning)
• Build Topology