calculating a map quality metric for ddm

Upload: luis-alberto-izarra

Post on 03-Apr-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    1/23

    Measuring Map Quality

    Material & Presentation by:

    Richard Frank

    Simon Fraser University

    February 2004

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    2/23

    Presentation Overview

    Motivation

    Uses of Map Quality

    Requirements

    Assumptions

    Definitions

    Algorithm Details

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    3/23

    Motivation

    Maps are generated in different ways Carefully by a human designer

    Automatically by a professional program Microsoft MapPoint

    Automatically by a free service www.mapquest.com

    They can also be shown on a wide variety of

    medium Due to resolution constraints, objects will

    change or disappear.

    http://www.mapquest.com/http://www.mapquest.com/
  • 7/28/2019 Calculating a Map Quality Metric for DDM

    4/23

    Motivation

    In printed form

    (map)

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    5/23

    Motivation

    www.MapQuest.com

    through a web-

    browser

    http://www.mapquest.com/http://www.mapquest.com/
  • 7/28/2019 Calculating a Map Quality Metric for DDM

    6/23

    Motivation

    Microsoft

    MapPoint(Standalone

    Program)

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    7/23

    Motivation

    Displayed on a

    PDA(Mapopolis)

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    8/23

    Uses of Map Quality

    Give the user some indication of howaccurate different aspects (location,shape, etc) of the map are

    Beneficial in providing the end-user amap that is much better tailored to theirspecific wants

    If the end user is interested in the structureof the maps, the computer can select thebest map out of a set of possible maps withbest possible structure

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    9/23

    Uses of Map Quality

    Can compare qualities of two alternate maps at

    same scale

    Can measure quality after the generalizationoperator, or after the visualization operator

    On the backside, it can be used to determine

    which data-cubes to generate

    Ones that can quickly produce, withoutgeneralization, on-demand maps above a certain

    quality

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    10/23

    Comparison

    compare proposed

    map to original map

    (the best possible

    map)

    To determine best

    alternative, compare

    measures of themaps

    Map Quality Indicator

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    11/23

    Requirements for good

    measurement

    Measure must take into account

    individual objects on a map

    the structure between them their distribution on a map

    These are enough to describe changes

    on a map No such measure currently exist

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    12/23

    Assumptions

    No symbolic representation for shapes

    Shapes remain shapes

    Were not concerned about changes inreadability

    Objects with holes are treated as multiple

    objects, i.e.: holes are treated as objectsthemselves

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    13/23

    Definition Voronoi Diagram

    Given a map of objects

    Find closest object or object edge

    If the closest edges belong to two or more objects which are

    equally close, then it is a Voronoi boundary

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    14/23

    Definition Voronoi Skeleton

    If the point is inside the object and the closest edges belong to

    two or more edges of the same object then it is part of the

    Voronoi Skeleton

    Voronoi Skeleton (in Red)

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    15/23

    Algorithm Components

    Object Shape Similarity

    Structure Similarity

    Information Content Similarity

    Each will generate a measure

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    16/23

    Shape Similarity

    A map is a collection ofobjects, which aftergeneralization canchange in shape

    The information lossduring the shape-change has to bemeasured

    Use: Edit-Distance of

    Voronoi Skeleton Idea adapted from Edit-

    Distance of ShockGraphs

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    17/23

    Shape Similarity

    Objects that contain holes are treated as multiple objects

    Small perturbations do not affect the Voronoi Skeleton

    Ideal for maps and bitmap objects

    Calculate edit distance by assigning costs totransformations that are required to change one

    structure into the other

    Object from Original Map Object from Generalized Map

    No Bump!

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    18/23

    Structure similarity

    Objects will be displacedduring generalization

    the position of an object willchange

    relative to the mapboundaries

    Relative distance to otherobjects

    Procedure

    Measure distances

    Input distances into matrix

    calculate a cosine similarity(standard way of comparingmatrices)

    Objects & their Voronoi Regions

    Before After

    Length between neighbors

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    19/23

    Information Content Similarity

    During generalization, several objects could bemerged/aggregated into one larger object, or can bedeleted

    There is loss of information because we looseinformation about the individual objects

    Loose 4 small objects Gain 1 large object

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    20/23

    Information Content: Entropy

    Usual method: Entropy Original calculation: SUM(Pi*ln(Pi))

    Should modify it by weighing objects according to

    the area of their Voronoi regions If information is lost when something disappears, the

    objects remaining become more important/influential

    Modified method: VE=SUM(Pi*ln(Pi)*%V)

    %V is the area of the Voronoi region for the objectdivided by the total map area

    Where Pi = Ki/N Ki = # of objects of type i

    N = total # of objects on the map

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    21/23

    Algorithm Components

    Consolidate the 3 measures into one number (representing thequality of the map)? Q = W1 * M1 + W2 * M2+ W3 * M3

    Where

    Q = Map quality measure Pi= some weight for metric i

    Mi= Measure of metric i

    The parameters can either be pre-defined, representing an idealsituation (if there is one), or can be left up to the user to let themspecify which issue is more important to them.

    OR

    Display all three resulting measures independently to the userand let them interpret the results

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    22/23

    Future Work

    Currently working on implementation

    Spatio-Temporal Data mining

    We can compare sub-areas of two mapsfrom different time periods to find area with

    most change, with possibility of restricting to

    any class

    ex: Find square kilometer with most road

    development

  • 7/28/2019 Calculating a Map Quality Metric for DDM

    23/23

    References

    Shape matching using edit-distance: an implementation (2001), Philip N.Klein, Thomas B. Sebastian, Benjamin B. Kimia, Symposium on Discrete

    Algorithms

    Framework for Matching shock graphs, Thomas B. Sebastian, Philip N.Klein, Benjamin B. Kimia,www.lems.brown.edu/vision/researchAreas/ShockMatching/shock-matching.html,

    10/16/2003 Quantitative measures for spatial information of maps, Zhilin Li and Peizhi

    Huang, Hong Kong Polytechnic University, Dec 2001

    Supporting Multiple Representations with Spatial Database ViewsManagement and the concept of VUEL, Yvan Bedard and Eveline Bernier,Universite Laval

    Fast computation of Generalized Voronoi Diagrams using GraphicsHardware. Kenneth E Hoff, Tim Culver, John Keyser, Ming Lin, Dinesh Manocha.

    University of North Carolina Voronoi Diagrams of Polygons: A Framework for shape representation. Niranjan

    Mayya & V.T. Rajan, University of Florida

    Conflict Reduction in Map Generalization using Iterative Improvement, JMark Ware & Christopher B. Jones, University of Glamorgan. 1998

    http://www.lems.brown.edu/vision/researchAreas/ShockMatching/shock-matching.htmlhttp://www.lems.brown.edu/vision/researchAreas/ShockMatching/shock-matching.htmlhttp://www.lems.brown.edu/vision/researchAreas/ShockMatching/shock-matching.htmlhttp://www.lems.brown.edu/vision/researchAreas/ShockMatching/shock-matching.html