Advanced Graphics Curves

download Advanced Graphics Curves

If you can't read please download the document

description

A lecture on curves in computer Graphics at ASU

Transcript of Advanced Graphics Curves

  • Advanced Computer Graphics I

    Yalin Wang

    Computer Science and Engineering,Arizona State University

    Sep. 01, 2015

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 1 / 34

  • Outline

    1 Motivation

    2 Curve Denitions

    3 Arc Length Parametrization

    4 The Frenet Frame

    5 Middle Axis

    6 Applications

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 2 / 34

  • Motivation

    Curve Analysis to Understand Geometry Shapes

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 3 / 34

  • Motivation

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 4 / 34

  • Motivation

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 5 / 34

  • Quiz

    Figure: How can we discriminate these four curves?

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 6 / 34

  • Denition

    Jordan Curve

    A (parametrized plane) curve is a continuous mapping m : I ! R2,where I = [ a;b] is an interval.

    The curve is closed if m(a) = m(b).

    A curve is a Jordan curve if it is closed and m has noself-intersection: m(x) = m(y) only for x = y or x ;y = a;b.

    The curve is piecewise C1 if m has everywhere left and rightderivatives, which coincide except at a nite number of poin ts.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 7 / 34

  • Denition

    Regular Curve

    A C1 curve m : I ! R2 is a regular curve if _mu 6= 0 for all u 2 I. If mis only piecewise C1, we extend the denition by requiring that allleft and right derivatives are non-vanishing.

    For the study of curve, it is essential that the curve is regular.

    Singular point: where _m(t) = 0.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 8 / 34

  • Parameterized Curve

    Denition

    An (innitely) differential map a : I ! R3 of an open interval I = ( a;b) ofreal line R into R3.

    a (t) = ( x(t);y(t);z(t)) .

    Tangent vector: _a (t) = ( x0(t);y0(t);z0(t)) .

    Trace: the image set a (I) R3.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 9 / 34

  • Parameterized Curve

    RemarksThe map a needs not to be one-to-one.

    a is simple if the map is one-to-one.

    Distinct curves can have the same trace:

    a (t) = ( cos(t);sin(t)) (1)

    b (t) = ( cos(2t);sin(2t)) (2)

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 10 / 34

  • Arc Length

    Denition

    Given t 2 I, the arc length of a regular curve a : I ! R3, from the pointt0, is

    s(t) =Z t

    t0j _a (t)jdt

    where

    j _a (t)j =q

    (x0(t))2 + ( y0(t))2 + ( z0(t))2

    is the length of the vector _a (t).

    Since _a (t) 6= 0, s(t) is a differentiable function of t , and dsdt = j _a (t)j.

    If the curve is arc length parameterized, then dsdt = 1 = j _a (t)j.

    Conversely, if j _a j 1, then s = t t0.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 11 / 34

  • Curves Parameterized by Arc Length

    Denition

    Let a : I ! R3 be a curve parameterized by arc length s 2 I, thenumber j a j = k is called the curvature of a at s.

    t(s) = _a (s)j _a (s) j.

    At point where k (s) 6= 0, the normal vector n(s) in the direction of_a (s) is well dened by a (s) = k (s)n(s).

    The plane determined by _a (s) and n(s) is called the osculatingplane.

    Binormal vector: b(s) = t(s) n(s).

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 12 / 34

  • The Frenet Frame

    Denition

    A curve g : (0;1) ! R3 is called a regular curve, if every component is aC funciton and for all t 2 (0;1); j dgdt j > 0.We have following denitions:

    t(s) = _g(s)

    n(s) =1k

    _t(s)

    b(s) = t(s) ^ n(s)

    (3)

    Therefore, there is an orthonormal frame, the Frenet frame, movingalong the curve:

    f g(s) : t(s);n(s);b(s)g:

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 13 / 34

  • The Frenet Frame

    Illustration

    Figure: Frenet frame for a point on a space curve

    DemoFrenetFrame-source.nb.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 14 / 34

  • Curvature

    Denition

    k (t) =jj _a (t) a (t)jj

    jj _a (t)jj3

    Osculating Circle

    Its radius is r = 1=k and its center is

    c(t) = a (t) + r (t)n(t)

    DemoCircleOfCurvature-source.nb.

    OsculatingCircles3D-source.nb.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 15 / 34

  • Osculating Circle

    Illustration

    Figure: An osculating circle.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 16 / 34

  • Parametric Curve Example: A Straight Line

    Straight LineGiven two points, a and b.

    The line connecting both points is p = tb + ( 1 t)a, t 2 [0;1].

    a

    b

    pt

    1-t

    Figure: A simple parametric curve example.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 17 / 34

  • Parametric Curve Example: Cubic B ezier Curves

    DenitionCubic B ezier curves

    x(t) = B30b0 + B31b1 + b

    32b2 + B

    33b3

    where where Bni (t) =

    ni

    (1 t)n i t i , with

    ni

    =

    (n!

    i !(n i)! if 0 i n0 else:

    DemoConstructionOfABezierCurve.cdf

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 18 / 34

  • Curvatures for B ezier Curves

    B ezier CurvesAt t = 0, we have

    k (0) = 2n 1

    narea[b0;b1;b2]

    jjb1 b0jj3

    At t = 1, we have

    k (1) = 2n 1

    narea[bn 2;bn 1;bn]

    jjbn bn 1jj3

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 19 / 34

  • Signed Curvature

    Signed Curvature

    k (t) =det [ _x(t); x (t)]

    jj _x(t)jj

    Demo: Evolutes of Some Basic Curves-source.nb.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 20 / 34

  • Curvature Tools

    max: 0.002

    min: -0.038

    Figure: A curve with its curvature plot

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 21 / 34

  • Curvature Tools

    max: 0.000

    min: -0.035

    Figure: An improved curve with its curvature plot

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 22 / 34

  • Middle Axis

    DenitionThe medial axis transform associates a skeleton-like structure to ashape.More precisely, represent a shape by an open connected bounded setin the plane,denoted . The skeleton of , denoted () , is the set ofall p such that, open disc B(p; r ) (where p 2 R2 is the center and radiusr > 0), is maximal in for some r > 0, i.e., () is the set of loci of thecenters of maximal discs.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 23 / 34

  • 2D Example

    Figure: Shapes with horizontal medial axes. The shapes are obtained with2uu(r

    2) = 2f , f is in the left column and the curves are in the right one.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 24 / 34

  • 2D Example

    Figure: Shapes generated from a medial axis with three linear branches.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 25 / 34

  • 3D Example

    Figure: Medial axis is used to analyze hippocampal morphometry.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 26 / 34

  • Cell Morphometry

    segmented lymphocytes0 Video frame grabslymphocyte

    T X (t, s) : [0, 2 ) [0, T ] R2extracted curve sequence

    A

    B CX (t)normalX (t)abnormalFigure: Curve analysis to model dynamic cellular morphology.

    X. An, Z. Liu, Y. Shi, N. Li, Y. Wang, S. Joshi. Modeling Dynamic CellularMorphology in Images, MICCAI 2012, Part I. 340-347

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 27 / 34

  • Cell Morphometry

    A1

    A2A3

    A4

    A5

    A6A7

    A8

    A9 A10

    A11

    A12

    A13

    A14

    A15

    A16

    A17

    A18

    A19

    A20

    A21

    N22

    N23N24

    N25

    N26

    N27

    N28

    N29N30

    N31N32

    N33

    N34N35N36

    N37N38N39N40

    N41

    N42

    MDS plot of distances between sequences (DTW)

    Dimension 1

    Dim

    ensi

    on 2

    A1

    A2

    A3

    A4

    A5

    A6

    A7 A8

    A9

    A10

    A11

    A12

    A13

    A14

    A15A16

    A17A18

    A19A20

    A21

    N22N23N24

    N25N26

    N27N28

    N29

    N30

    N31N32

    N33

    N34

    N35N36

    N37N38N39N40

    N41N42

    MDS plot of distances between sequences (no DTW)

    Dimension 1

    Dim

    ensi

    on 2

    Figure: Shape space illustration. Projections of the pairwise distancesbetween 42 sequences without dynamic time warping (DTW) (left) and withDTW (right).

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 28 / 34

  • Conformal Welding

    SketchSuppose = g0;g1; :::;gk are non-intersecting smooth closedcurves on the complex plane.

    partitions the plane to a set of connected components 0; 1; :::; s , each segment i is a singly-connected domain.

    Construct f k : k ! Dk to map each segment k to a circledomain Dk , 0 k s.

    Assume gi 2 = j \ k , then f j (gi ) is a circular boundary in Djwhile f k (gi ) is another circle on Dk .

    Let fi = f j f 1k : S

    1 ! S1 be the diffeomorphism from the circle toitself. It is called a signature of gi .

    ReferenceE. Sharon and D. Mumford. 2D-shape analysis using conformalmapping. Int. J. Comput. Vision, 70:55-75, October 2006

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 29 / 34

  • 2D Conformal Wedling

    Figure: Conformal disparity between the interior and exterior of three planarcurves.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 30 / 34

  • 2D Conformal Wedling Application

    Figure: Planar signature.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 31 / 34

  • System Overview

    Figure: Diffeomorphism signature via uniformization mapping.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 32 / 34

  • 3D Shape Signatures

    Figure: 3D shape signatures by conformal welding.

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 33 / 34

  • Experimental Results

    Figure: Distribution of shape signatures (12 control vs 12 Alzheimers diseasepatients).

    Y. Wang (ASU) Shape Analysis Sep. 01, 2015 34 / 34

    MotivationCurve DefinitionsArc Length ParametrizationThe Frenet FrameMiddle Axis