DTM Generation From Analogue Maps

1
DTM Generation From Analogue Maps

• By
• Varshosaz

2
Using cartographic data sources

• Data digitised mainly from contour maps
• Digitising contours leads to oversampling over
  the contours and undersampling between the
  contours
• Errors are inherent in paper maps due to drawing,
  generalisation, reproduction, etc.
• May still be cost effective at medium or small
  scale with national coverage

3
Using cartographic data sources (cont.)

• Digitisation can be
• Manual line following
• Semi or fully automatic line following
• Automatic raster scanning and vectorisation

4
Manual digitising
5
Converting a paper contour map into vector format

• Scanning
• Image type TIF, GIF,BMP, etc.
• Resolution
• Removing noise
• Median filter
• Neighbourhood averaging
• Contour detection
• Edge detection
• Binary extraction
• Skeletonisation
• Vectorisation (Contour Following)

6
Scanning Maps

• The scanning process converts the analogue
  (paper) maps into raster (digital) format.
• Selection of an appropriate dpi for the scan is
  in essence the determining factor of how many
  dots per inch the scanner will record.
• Limitations
• Scanning resolution of the scanner itself
• Hardware issues and image file sizes.

7
Scanners

• Accuracy
• Photogrammetric
• Desktop Publishing
• Mechanism
• Flatbed
• Drumbed

8
Mechanism of Scanners
9
Photogrammetric Scanners

• Stable and Known Geometry
• Accurate
• Expensive
• Limited Availability

10
DeskTop Publishing Scanners

• Everyday scanners
• Cheap
• High availability
• Low accuracy and large distortions

11
DTP Distortions
12
DTP Distortion Removal
13
Noise Removal

• Noise is present in any scanned map due to
• Poor-sampling process
• Poor original map.
• Objective remove unwanted noise
  before detecting, binarizingand vectorizing the
  contours.
• Principle Applying spatial domain smoothing
  techniques in local neighborhoods of the scanned
  map (image).

14
Noise Removal Median Filter

• Sorting the intensity values in ascending or
  descending order.
• Choose the median as new centre value.
• Characteristics
• Removes pixels in the neighborhood that are
  dramatically different (noise) from the rest.
• It does preserve sharpness of an image.

15
Noise Removal Median Filter
16
Noise Removal Neighborhood Averaging
17
Contour Detection

• Scanned contours are linear features,
• They are bounded by edges (the transition or
  boundary between the contours and the
  background).
• An edge is a discontinuity in the two dimensional
  grey scale function.
• Abrupt change in the gray level intensity within
  an area of the image space constitutes an edge.
• Contour detection (edge detection) refers to the
  process that examines the scanned map for
  discontinuities in the grey level function.

18
Edge Detection

• Edges are characterized by discontinuities in the
  gray values at their location.
• A typical edge detection algorithm uses first
  derivative of an image eg Sobel

19
Original Image
20
Detected Edges
21
Edge Detection SOBEL Filter (1968)
22
Binary Extraction

• Objective Reduce scanned map resolution from 256
  intensities to two intensities.
• Reduces the scanned map into two categories
  Contours and Background.
• Applied to maps (images) that have been
  adequately enhanced, smoothed, and the contours
  have been detected as edges.

23
Threshold Binary Extraction
24
Edge Detection Binarization
25
Skeleton Processing (Thinning)

• Gradient filtering and the binarizationproduce
  edges wider than one pixel.
• The required final position of the edge lies
  roughly in the middle of this wider edge.
• Extracting the center position of the edge is
  known as skeleton processing.
• Based on a square array of image (3x3, 5x5, 7x7,
  etc).
• Note that as the template size increases, the
  number of different combinations dramatically
  increases and so does the computation time.

26
Skeleton Processing
Consider this 3x3 approach which produces a
skeleton line which is close to a medial line.
27
Skeleton Processing
28
Skeleton Processing
29
Contour Following

• Output of Skeleton Processing
• Thin contour lines with one pixel width in the
  area of interest.
• To extract the whole contour, we need to trace
  pixels and obtain their positions.
• The vectorization processes usually done in
  semiautomatic mode, where the operator provides
  the initial points.

30
Contour Following

• The initial direction can either be given by the
  operator or be determined through the automatic
  search procedure.
• In the latter case, the initial direction is
  actually approximated as 0 degrees (i.e.,
  pointing upward in the scanned map).
• User usually defines the number of search
  directions.
• Example define the initial direction as 0 degree
  and the number of directional matrices as 13.

31
Contour Following