logo

SCIENTIA SINICA Informationis, Volume 47, Issue 4: 428-441(2017) https://doi.org/10.1360/N112016-00172

Spherical parameterization based on planar ARAP+ method

More info
  • ReceivedJul 14, 2016
  • AcceptedNov 28, 2016
  • PublishedFeb 22, 2017

Abstract

This paper proposes a novel spherical parameterization to process the genus zero models. The method is based on the planar ARAP+ method and further research of the local/global parameterization. It consists of two main steps, local phase and global phase. In the local phase, we employ the planar ARAP+ method to optimize the Spring energy, and achieve the displacement of planar vertex around its 1-ring neighborhood, then map the new vertex to the 3D 1-ring neighborhood on the sphere. In the global phase, we add the spherical constrains to the planar ARAP+ method, and then obtain the final result iteratively according to the Newton method. Numerical results demonstrate that our method is efficient and convergent, and outperforms several popular methods in term of controlling the distortion measures (angle, area, rigidity). Furthermore, it achieves a better visualization performance in texture mapping.


Funded by

国家自然科学基金(61432003,61572105,11171052,61328206)


Acknowledgment

我们特别感谢辽宁师范大学的亓万锋 博士和大连理工大学的孟兆良老师在本文研究过程 中提出的许多宝贵意见, 同时还要衷心感谢以色列理工学院的 Emil Saucan 教授和纽约州立大学的顾险峰教授与我们保持密切的科研合作.

  • Figure 1

    (Color online) The five types of spherical parameterization. (a) Convex; (b) ARAP $^{[4]}$; (c) BLD [9]; (d) our conformal; (e) our isometric

  • Figure 2

    (Color online) Original mesh and initial parameterization

  • Figure 3

    The processing of the local 1-ring neighborhood. (a) Local flattening of original mesh;(b) local optimization;(c) local flattening of initial mesh;(d) global solution;(e) spherical projection;(f) spherical parameterization

  • Figure 4

    (Color online) The iterative results produced by our method. (a) Original mesh; (b) 1 iteration; (c) 2 iterations; (d) 3 iterations

  • Figure 5

    (Color online) The post-processing result. (a) Original mesh; (b) before processing; (c) after processing

  • Figure 6

    (Color online) The spherical parameterization and texture mapping. (a) Our conformal; (b) our isometric; mbox(c) our authalic

  • Figure 7

    (Color online) The numerical convergence. (a) Our conformal; (b) our isometric; (c) our authalic

  • Figure 8

    (Color online) Texture mapping of the multi-boundary model. (a) Original mesh; (b) planar parameterization; (c) planar texture mapping; (d) spherical parameterization; (e) cube; (f) spherical texture mapping

  • Figure 9

    (Color online) Comparison of spherical parameterization (ARAP: the first and third rows of (a)–(c); ARAP+: the second and fourth rows of (a)–(c)) and texture mapping of ARAP and ARAP+. (a) 1 iteration; (b) 2 iteration; (c) 3 iteration; (d) texture mapping of ARAP; (e) texture mapping of ARAP+

  • Figure 10

    (Color online) Comparison of spherical parameterization and texture mapping. (a) Convex; (b) ARAP; mbox(c) our conformal;d) our isometric; (e) our authalic

  • Figure 11

    (Color online) The processing of high-curvature model. (a) Original mesh;(b) our conformal; (c) our isometric; (d) our authalic

  •   

    Algorithm 1 基于平面 ARAP+ 方法的球面参数化算法

    Require:A 3D mesh $S$;

    Output:A spherical mesh $S^{\ast}$;

    $S^*=S_0^*$; // 初始化ŁOOP

    for $i = 1 : n$

    // 局部优化

    $\omega_{i,j}$ $\Leftarrow$ ComputeWeights(RingNodes($p_i$));

    RingNodes($p^{\prime}_i$) $\Leftarrow$ ComputeWeights(RingNodes($p_i$));

    RingNodes($q^{\prime}_i$) $\Leftarrow$ ComputeWeights(RingNodes($q_i$));

    $L_{(i,j,j+1)}$ $\Leftarrow$ FittingMatrix($\triangle p^{\prime}_ip^{\prime}_jp^{\prime}_{j+1}$, $\triangle q^{\prime}_iq^{\prime}_jq^{\prime}_{j+1}$);

    New($q^{\prime}_i$) $\Leftarrow$ Relocate2D($\triangle p^{\prime}_ip^{\prime}_jp^{\prime}_{j+1}$, $\triangle q^{\prime}_iq^{\prime}_jq^{\prime}_{j+1}$);

    $\widetilde{\omega}_{i,j}$ $\Leftarrow$ MeanValue(RingNodes(New($q^{\prime}_i$)));

    $q^{\prime\prime}_i$ $\Leftarrow$ Relocate3D($\widetilde{\omega}_{i,j}$,RingNodes($q_i$));

    $B_i$ $\Leftarrow$ MotionVector($q^{\prime\prime}_i$, $\widetilde{q}_i$);

    if $\|B_i\|<10^{-3}$, then

    $B_i=(0,0,0)$; // 后处理

    end if

    end for

    $S^*_1$ $\Leftarrow$ ComputePara($\omega$, $B_i$); // 全局求解

    if Iteration $<$ 5, then

    $S^*_0=S^*_1$;

    else

    $S^*=S^*_0$;

    break;

    end if

    ENDLOOP

  • Table 1   Comparison of the distortion measures and running time between planar ARAP and ARAP+
    Method $n$ Child tiny (V:6368, F:12732) Skull tiny (V:5007, F:10010)
    Angle Area Rigidity Time (s) Angle Area Rigidity Time (s)
    1 2.28 6.42 2.93 10.66 2.03 5.65 1.59 6.84
    ARAP 2 2.52 6.23 2.85 21.69 2.11 4.89 1.58 13.85
    3 3.41 5.75 2.79 32.45 2.15 3.84 1.56 21.30
    1 2.24 6.13 2.89 10.62 2.02 5.19 1.57 6.91
    ARAP+ 2 2.45 5.69 2.82 21.59 2.05 4.52 1.55 13.95
    3 3.26 5.09 2.75 32.39 2.13 3.5 1.53 21.43
  • Table 2   The comparison of five methods on the distortion measures and running time
    Method Man tiny (V:5785, F:11566) Girl tiny (V:6610, F:13216)
    Angle Area Rigidity Time (s) Angle Area Rigidity Time (s)
    Convex 2.13 9.45 2.35 16.2 2.03 9.11 2.68 17.3
    ARAP 2.31 3.63 1.73 20.4 2.34 4.04 2.15 22.4
    Our conformal 2.07 9.61 2.16 17.4 2.01 7.65 2.73 19.1
    Our isometric 2.29 3.54 1.63 17.1 2.15 3.71 1.93 18.6
    Our authalic 2.94 3.46 4.04 19.6 2.81 3.41 4.55 21.5
    Method Kitten tiny (V:10200, F:20396) Canonical tiny (V:2243, F:4482)
    Angle Area Rigidity Time (s) Angle Area Rigidity Time (s)
    Convex 2.11 3.37 0.82 63.9 2.01 4.65 2.14 2.9
    ARAP 2.15 2.62 0.58 72.1 2.11 3.35 1.65 4.8
    Our conformal 2.03 3.65 0.86 67.7 1.98 4.59 2.12 3.4
    Our isometric 2.12 2.56 0.54 66.4 2.09 3.32 1.61 3.2
    Our authalic 2.54 2.51 1.47 70.2 3.34 3.29 2.51 4.3

Copyright 2020 Science China Press Co., Ltd. 《中国科学》杂志社有限责任公司 版权所有

京ICP备18024590号-1       京公网安备11010102003388号