SlideShare a Scribd company logo
[Garland99] Thesis
[Garland98] Simplifying Surfaces with Color and Textures using Quadric Error Metrics
[Garland97] Surface Simplification using Quadric Error Metrics

Flip check – p.56




Homogeneous Variant – p.50




The homogeneous form of the quadrics can be transformed by a linear transform.
However this formulation is less convenient and less efficient (4x4 operations).

Vertex Placement Policies – p.51
Taking partial derivatives, we see that the gradient of Q is




and its error is
These are instances of the minimum of a positive definite quadratic form. Geometrically
it can be interpreted as the least-square optimal point which best fits the set of planes
represented by the quadric. It lies in the center of the ellipsoidal isosurfaces of Q.

Discontinuities and Constraints
The basic algorithm ignores boundary curves. We form a plane perpendicular to a
boundary; calculate a quadric add a large penalty and add it to both ends of the edge.
When using area-weighted quadrics, the constraints must be properly weighted. The
obvious choice of weighting by adjacent face’s area will led to dependency on the
tessellation next to the boundary. A better choice is to use the squared length of the
boundary edge. The same can be applied to ‘feature edges’ within a mesh.

We can also formulate a constraint for points too.

Volume Quadric and relation to Lindstrom




Using simple vector algebra we can get the squared volume of the tetrahedron as




m is the cross-prod of the edges. (4.9) is equivalent to scaling the fundamental quadric by
w2/9 where w is the area of the contributing face T.

Principal Components of Quadrics

We can regard A as the covariance of the normals with mean 0. The eigenvectors should
be roughly be in the direction of the average normal, direction of maximum curvature and
direction of minimum curvature.

Ambiguity of Sharp Angles
The initial quadric assigned to both of these cases are the same since its incident
segments lie along the same lines. In (b) the curve will become a single spike and quadric
will allow points to move freely along the line.

[Kho03] User-Guided Simplification
The user can adaptively weight areas of the models. The actual weight applied is wlogV
where V is the number of vertices in the input in order to get similar results of a feature
from two different scales of the model. Using weights leaves more polygons in areas
where user feel is important. Another tool is to use constraints – like the virtual plane in
the Garland’s original work. To avoid constraints unnecessary increase cost we keep the
constraint quadrics separate from the geometric ones and only add them when we are
computing the optimal positions. They propose 3 types of constraints – contour, plane
and point.

[Lindstrom98] Fast and Memory Efficient Polygonal Simplification

They pioneered the memoryless approach whereby the cost is computed wrt the current
simplified meshes and not the original mesh. C(v) as a sum of squared tetrahedral
volumes formed by ‘v’ and the faces around the vertex neighborhood of ‘v’ in the
simplified mesh.

The basic approach to finding the optimal position is to combine a number of linear
equality constraints aiTv = bi, i.e. v is the intersection of three non-parallel planes in 3.
If two of more of these planes are nearly parallel, minor perturbations to the plane
coefficients lead to large variations in the solution. So we add a constraint to a set of
existing constraints only if the plane normal does not fall within an angle of all linear
combinations of the plane normals of the previous constraints. See [Lindstrom00] for a
SVD based alternative.

Volume Preservation
By considering the signed changes in the volumes of the tetrahedrons formed by the new
vertex and the old triangle.




where ni is the outward normal vector of triangle ti, with magnitude twice the area of ti.

Boundary Preservation
For a planar boundary, we attempt to preserve the area enclosed by the boundary. We use
signed changes in area.

Volume Optimization
The volume preservation constraint leaves an entire plane of candidate vertices. To
further constrain the vertex position we minimize the unsigned volume of each individual
tetrahedron, which is a measure of the local surface error for each triangle in the edge’s
neighborhood. If all of these are coplanar the volume optimization has many solutions

Boundary Optimization
We minimize the sum of squared areas.

Triangle Shape Optimization




Which is the sum of square of the edge lengths incident upon v. This unsure the area to
perimeter ration is maximized. This is used as a last resort only when both fV and fB are
both close to zero.

Edge Costs




[Lindstrom00] Out-of-core simplification of large polygonal meshes
Points are binned into grid cells, within each cell qem is used to collapse vertices and to
calculate the new vertex. Only a single pass is needed through the input model. The qem
used is the memoryless version from [Lindstrom98]. A hash table based on the grid cell
index is used to store the input. The quadric for the triangle is computed and added to all
3 bins. Each bin represents one output vertex.




Where n is the 4-vector made up of area-weighted triangle normal and scalar triple
product of its vertices. Since Q is symmetric and c is not used, only 9 scalars are needed.
After adding all the quadric within a grid, we use the block decomposition above and
solve the Ax = b for the optimal position. They introduce a robust SVD based way to
invert the quadric matrix instead of the linear constraints of [Lindstrom98].
[Hoppe] New Quadric Metric for Simplifying Meshes with Appearance Attributes

They confirmed the memoryless method of Lindstrom gives better quality. In the
standard QEM scheme the Qv are computed on the original mesh and subsequently
summed. As a result, the merging of the non-parallel ovals (corresponding to fine level
features) gives rise to tight spherical quadrics that lock vertices and prevent further
simplification, even though the resulting mesh is planar.




Garland metric is quadratic in space – i.e. for m attributes it need o(n^2) space.

The new quadric defines both the geometric error and attribute error based on geometric
correspondence in 3D. Rather than projecting the point p onto the mesh face in an
abstract higher-dimensional space R3+m. The error is defined as the sum of the geometric
error and the attribute error. The attribute error is the squared deviation between s and
the value s’ interpolated from face f at the projected point p’. The projection is done in
R3. The new metric takes 11 + 4m coefficients, which is linear in m.




In the memoryless scheme the Qf are computed on the face neighborhood of the
collapsing edge. The memoryless simplification makes storing QEM’s unnecessary. For
speed it is still good to stored the values (area(f).Qf(v)) on the faces.

The new QEM sometimes shrinks the model geometry in areas of high attribute gradient
– new vertex is pushed towards the center of curvature of the surface at sharp attribute
transitions (see fig.7 above).

Preserving volume is equivalent to a linear constraint gTvol p + dvol = 0. It is solved using a
Lagrange multiplier. The volumetric gradient gvol is the sum of the face normals of Fi+1
weighted by 1/3 of their face areas.


[Hoppe00] Efficient Minimizatio of New Quadric Metric for Simplifying Meshes with
Appearance Attributes
The minimization of the quadric metric with m appearance attribute involves solving a
linear system of size (3+m)x(3+m). The system has only O(m) nonzero entries, so it can
be solved in O(m2) using sparse solvers such as conjugate gradients. Here the authors
show the special structure of the sparsity permits the system to be solved in O(m) time.
[DeCoro07] Real Time Simplification using the GPU
They use [Lindstrom00] qem based vertex collapse as the base and add a probabilistic
octree instead of a grid. The basic alg is very simple – it use a user specified uniform grid
and use the vertex shader to compute the grid index. The index is used to accumulate a
per-cluster qem. In pass 2 the qem is solved using a matrix invert to get the optimal vertex
position. In pass 3, the input mesh is send through the a 2 nd time, geometry shader is used
to cull triangles are collapsed; otherwise the shader retrieve the positions from pass 2 and
streamed out into a GPU buffer.

Two warping function are introduced a) the world-view projective transform b) area-of-
interest by a Gaussian. It is inverted using the Gaussian error function. The inverse
warping function is used only when computing the cluster coordinates.

To support adaptive resolution a probabilistic octree using hash was introduced.

[Frey99] Surface Mesh Quality Evaluation
Measures the aspect ratio of the set of triangles to define the quality.
              1                        6 Sf
 K Tri (M j )       Q f , where Q f               . Sf is the area of the face f, pk is the half-
              F f F                     3 p k .hk
perimeter of f and hf is the longest edge.

[Southern00] Evaluation of Memoryless Simplification
They evaluated the common alternatives for vertex placement – a) 0 bit: mid-point b) 1
bit: half-edge collapse c) 2 bits: to encode the end-pts + mid-pt.

They defined a triangle quality metric that is faster than [Frey99]
                               3

               F
                                    e ij
           1                  j 1
K (M )               (                       ) ; divide the edge length sum by 3 times the minimum edge.
           F   i 1       3(min 3j 1 e ij )
                                                                                       F
                                                                                   1
For volume, they use the Gaussian Divergence Formula K vol ( M )                             (v1i v2 ) v3 ,
                                                                                                   i    i

                                                                                   6   i 1

More Related Content

What's hot

Parallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-JoinsParallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-Joins
Jonny Daenen
 
Base Excited Systems
Base Excited SystemsBase Excited Systems
Base Excited Systems
Teja Ande
 
Retaining wall design_example
Retaining wall design_exampleRetaining wall design_example
Retaining wall design_examplealibunar
 
AS4100 Steel Design Webinar Worked Examples
AS4100 Steel Design Webinar Worked ExamplesAS4100 Steel Design Webinar Worked Examples
AS4100 Steel Design Webinar Worked Examples
ClearCalcs
 
Bezier curve
Bezier curveBezier curve
Bezier curve
amiyadash
 
Braking analysis of a four wheel
Braking  analysis of a four wheelBraking  analysis of a four wheel
Braking analysis of a four wheel
Pavan Narkhede
 
A multiplicative time series model
A multiplicative time series modelA multiplicative time series model
A multiplicative time series model
Mohammed Awad
 
Brake analysis of four wheel vehicle
Brake analysis of  four wheel vehicleBrake analysis of  four wheel vehicle
Brake analysis of four wheel vehicle
Nilraj Vasandia
 
Ch11 ssm
Ch11 ssmCh11 ssm
Ch11 ssm
Marta Díaz
 
2016 SMU Research Day
2016 SMU Research Day2016 SMU Research Day
2016 SMU Research DayLiu Yang
 
Colloquium presentation
Colloquium presentationColloquium presentation
Colloquium presentation
bgeron
 
An Analysis of Graph Cut Size for Transductive Learning
An Analysis of Graph Cut Size for Transductive LearningAn Analysis of Graph Cut Size for Transductive Learning
An Analysis of Graph Cut Size for Transductive Learningbutest
 
Proceedings of the institution of mechanical engineers, part g journal of ae...
Proceedings of the institution of mechanical engineers, part g  journal of ae...Proceedings of the institution of mechanical engineers, part g  journal of ae...
Proceedings of the institution of mechanical engineers, part g journal of ae...
Spectre Industries, LLC
 
Tipler para cientistas e engenheiros resolução 6 edição
Tipler para cientistas e engenheiros resolução 6 ediçãoTipler para cientistas e engenheiros resolução 6 edição
Tipler para cientistas e engenheiros resolução 6 edição
Lowrrayny Franchesca
 
Modern physics paul a. tipler 6ª edição solutio manual
Modern physics   paul a. tipler 6ª edição solutio manualModern physics   paul a. tipler 6ª edição solutio manual
Modern physics paul a. tipler 6ª edição solutio manual
Izabela Ferreira
 
Some fixed point theorems of expansion mapping in g-metric spaces
Some fixed point theorems of expansion mapping in g-metric spacesSome fixed point theorems of expansion mapping in g-metric spaces
Some fixed point theorems of expansion mapping in g-metric spaces
inventionjournals
 

What's hot (19)

Parallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-JoinsParallel Evaluation of Multi-Semi-Joins
Parallel Evaluation of Multi-Semi-Joins
 
Base Excited Systems
Base Excited SystemsBase Excited Systems
Base Excited Systems
 
modification of rmr system for indian coal mines
 modification of rmr system for indian coal mines modification of rmr system for indian coal mines
modification of rmr system for indian coal mines
 
Retaining wall design_example
Retaining wall design_exampleRetaining wall design_example
Retaining wall design_example
 
AS4100 Steel Design Webinar Worked Examples
AS4100 Steel Design Webinar Worked ExamplesAS4100 Steel Design Webinar Worked Examples
AS4100 Steel Design Webinar Worked Examples
 
Bezier curve
Bezier curveBezier curve
Bezier curve
 
Braking analysis of a four wheel
Braking  analysis of a four wheelBraking  analysis of a four wheel
Braking analysis of a four wheel
 
Problem and solution i ph o 27
Problem and solution i ph o 27Problem and solution i ph o 27
Problem and solution i ph o 27
 
A multiplicative time series model
A multiplicative time series modelA multiplicative time series model
A multiplicative time series model
 
Brake analysis of four wheel vehicle
Brake analysis of  four wheel vehicleBrake analysis of  four wheel vehicle
Brake analysis of four wheel vehicle
 
Ch11 ssm
Ch11 ssmCh11 ssm
Ch11 ssm
 
2016 SMU Research Day
2016 SMU Research Day2016 SMU Research Day
2016 SMU Research Day
 
Colloquium presentation
Colloquium presentationColloquium presentation
Colloquium presentation
 
An Analysis of Graph Cut Size for Transductive Learning
An Analysis of Graph Cut Size for Transductive LearningAn Analysis of Graph Cut Size for Transductive Learning
An Analysis of Graph Cut Size for Transductive Learning
 
Proceedings of the institution of mechanical engineers, part g journal of ae...
Proceedings of the institution of mechanical engineers, part g  journal of ae...Proceedings of the institution of mechanical engineers, part g  journal of ae...
Proceedings of the institution of mechanical engineers, part g journal of ae...
 
Tipler para cientistas e engenheiros resolução 6 edição
Tipler para cientistas e engenheiros resolução 6 ediçãoTipler para cientistas e engenheiros resolução 6 edição
Tipler para cientistas e engenheiros resolução 6 edição
 
Modern physics paul a. tipler 6ª edição solutio manual
Modern physics   paul a. tipler 6ª edição solutio manualModern physics   paul a. tipler 6ª edição solutio manual
Modern physics paul a. tipler 6ª edição solutio manual
 
Lect14
Lect14Lect14
Lect14
 
Some fixed point theorems of expansion mapping in g-metric spaces
Some fixed point theorems of expansion mapping in g-metric spacesSome fixed point theorems of expansion mapping in g-metric spaces
Some fixed point theorems of expansion mapping in g-metric spaces
 

Similar to Mesh simplification notes

Triple integrals and applications
Triple integrals and applicationsTriple integrals and applications
Validation of Polarization angles Based Resonance Modes
Validation of Polarization angles Based Resonance Modes Validation of Polarization angles Based Resonance Modes
Validation of Polarization angles Based Resonance Modes
IJERA Editor
 
Newton cotes integration method
Newton cotes integration  methodNewton cotes integration  method
Newton cotes integration method
shashikant pabari
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and Development
IJERD Editor
 
Template-Based Paper Reconstruction from a Single Image is Well Posed when th...
Template-Based Paper Reconstruction from a Single Image is Well Posed when th...Template-Based Paper Reconstruction from a Single Image is Well Posed when th...
Template-Based Paper Reconstruction from a Single Image is Well Posed when th...
pigei
 
graph_theory_ch_3_20201124.ppt
graph_theory_ch_3_20201124.pptgraph_theory_ch_3_20201124.ppt
graph_theory_ch_3_20201124.ppt
ArwansyahDipanegara
 
Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...
guest31063e
 
Subquad multi ff
Subquad multi ffSubquad multi ff
Subquad multi ff
Fabian Velazquez
 
Curves and surfaces
Curves and surfacesCurves and surfaces
Curves and surfaces
Mohammed Mahmoud
 
control engineering revision
control engineering revisioncontrol engineering revision
control engineering revision
ragu nath
 
membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...
membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...
membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...Florian Gauss
 
FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...
FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...
FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...grssieee
 
Chapter 4
Chapter 4Chapter 4
Chapter 4
Eka Puspita Sari
 
Curves
CurvesCurves
Curves
Yatin Singh
 
A Condensation-Projection Method For The Generalized Eigenvalue Problem
A Condensation-Projection Method For The Generalized Eigenvalue ProblemA Condensation-Projection Method For The Generalized Eigenvalue Problem
A Condensation-Projection Method For The Generalized Eigenvalue Problem
Scott Donald
 
HOME ASSIGNMENT (0).pptx
HOME ASSIGNMENT (0).pptxHOME ASSIGNMENT (0).pptx
HOME ASSIGNMENT (0).pptx
SayedulHassan1
 
Q1Perform the two basic operations of multiplication and divisio.docx
Q1Perform the two basic operations of multiplication and divisio.docxQ1Perform the two basic operations of multiplication and divisio.docx
Q1Perform the two basic operations of multiplication and divisio.docx
amrit47
 

Similar to Mesh simplification notes (20)

curve one
curve onecurve one
curve one
 
Triple integrals and applications
Triple integrals and applicationsTriple integrals and applications
Triple integrals and applications
 
Validation of Polarization angles Based Resonance Modes
Validation of Polarization angles Based Resonance Modes Validation of Polarization angles Based Resonance Modes
Validation of Polarization angles Based Resonance Modes
 
Newton cotes integration method
Newton cotes integration  methodNewton cotes integration  method
Newton cotes integration method
 
Lar calc10 ch04_sec2
Lar calc10 ch04_sec2Lar calc10 ch04_sec2
Lar calc10 ch04_sec2
 
International Journal of Engineering Research and Development
International Journal of Engineering Research and DevelopmentInternational Journal of Engineering Research and Development
International Journal of Engineering Research and Development
 
Template-Based Paper Reconstruction from a Single Image is Well Posed when th...
Template-Based Paper Reconstruction from a Single Image is Well Posed when th...Template-Based Paper Reconstruction from a Single Image is Well Posed when th...
Template-Based Paper Reconstruction from a Single Image is Well Posed when th...
 
graph_theory_ch_3_20201124.ppt
graph_theory_ch_3_20201124.pptgraph_theory_ch_3_20201124.ppt
graph_theory_ch_3_20201124.ppt
 
Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...Normalized averaging using adaptive applicability functions with applications...
Normalized averaging using adaptive applicability functions with applications...
 
Subquad multi ff
Subquad multi ffSubquad multi ff
Subquad multi ff
 
Curves and surfaces
Curves and surfacesCurves and surfaces
Curves and surfaces
 
L 4 4
L 4 4L 4 4
L 4 4
 
control engineering revision
control engineering revisioncontrol engineering revision
control engineering revision
 
membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...
membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...
membranes2015_Attractive Planar Panelization using Dynamic Relaxation Princip...
 
FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...
FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...
FR4.L09.5 - THREE DIMENSIONAL RECONSTRUCTION OF URBAN AREAS USING JOINTLY PHA...
 
Chapter 4
Chapter 4Chapter 4
Chapter 4
 
Curves
CurvesCurves
Curves
 
A Condensation-Projection Method For The Generalized Eigenvalue Problem
A Condensation-Projection Method For The Generalized Eigenvalue ProblemA Condensation-Projection Method For The Generalized Eigenvalue Problem
A Condensation-Projection Method For The Generalized Eigenvalue Problem
 
HOME ASSIGNMENT (0).pptx
HOME ASSIGNMENT (0).pptxHOME ASSIGNMENT (0).pptx
HOME ASSIGNMENT (0).pptx
 
Q1Perform the two basic operations of multiplication and divisio.docx
Q1Perform the two basic operations of multiplication and divisio.docxQ1Perform the two basic operations of multiplication and divisio.docx
Q1Perform the two basic operations of multiplication and divisio.docx
 

More from Manchor Ko

Non-Local Means and its Applications
Non-Local Means and its ApplicationsNon-Local Means and its Applications
Non-Local Means and its Applications
Manchor Ko
 
ParallelRandom-mannyko
ParallelRandom-mannykoParallelRandom-mannyko
ParallelRandom-mannykoManchor Ko
 
Parallel Random Generator - GDC 2015
Parallel Random Generator - GDC 2015Parallel Random Generator - GDC 2015
Parallel Random Generator - GDC 2015
Manchor Ko
 
Dictionary Learning in Games - GDC 2014
Dictionary Learning in Games - GDC 2014Dictionary Learning in Games - GDC 2014
Dictionary Learning in Games - GDC 2014
Manchor Ko
 
omp-and-k-svd - Gdc2013
omp-and-k-svd - Gdc2013omp-and-k-svd - Gdc2013
omp-and-k-svd - Gdc2013
Manchor Ko
 
Gdc2012 frames, sparsity and global illumination
Gdc2012 frames, sparsity and global illumination Gdc2012 frames, sparsity and global illumination
Gdc2012 frames, sparsity and global illumination
Manchor Ko
 
Cache aware hybrid sorter
Cache aware hybrid sorterCache aware hybrid sorter
Cache aware hybrid sorterManchor Ko
 

More from Manchor Ko (7)

Non-Local Means and its Applications
Non-Local Means and its ApplicationsNon-Local Means and its Applications
Non-Local Means and its Applications
 
ParallelRandom-mannyko
ParallelRandom-mannykoParallelRandom-mannyko
ParallelRandom-mannyko
 
Parallel Random Generator - GDC 2015
Parallel Random Generator - GDC 2015Parallel Random Generator - GDC 2015
Parallel Random Generator - GDC 2015
 
Dictionary Learning in Games - GDC 2014
Dictionary Learning in Games - GDC 2014Dictionary Learning in Games - GDC 2014
Dictionary Learning in Games - GDC 2014
 
omp-and-k-svd - Gdc2013
omp-and-k-svd - Gdc2013omp-and-k-svd - Gdc2013
omp-and-k-svd - Gdc2013
 
Gdc2012 frames, sparsity and global illumination
Gdc2012 frames, sparsity and global illumination Gdc2012 frames, sparsity and global illumination
Gdc2012 frames, sparsity and global illumination
 
Cache aware hybrid sorter
Cache aware hybrid sorterCache aware hybrid sorter
Cache aware hybrid sorter
 

Mesh simplification notes

  • 1. [Garland99] Thesis [Garland98] Simplifying Surfaces with Color and Textures using Quadric Error Metrics [Garland97] Surface Simplification using Quadric Error Metrics Flip check – p.56 Homogeneous Variant – p.50 The homogeneous form of the quadrics can be transformed by a linear transform. However this formulation is less convenient and less efficient (4x4 operations). Vertex Placement Policies – p.51 Taking partial derivatives, we see that the gradient of Q is and its error is
  • 2. These are instances of the minimum of a positive definite quadratic form. Geometrically it can be interpreted as the least-square optimal point which best fits the set of planes represented by the quadric. It lies in the center of the ellipsoidal isosurfaces of Q. Discontinuities and Constraints The basic algorithm ignores boundary curves. We form a plane perpendicular to a boundary; calculate a quadric add a large penalty and add it to both ends of the edge. When using area-weighted quadrics, the constraints must be properly weighted. The obvious choice of weighting by adjacent face’s area will led to dependency on the tessellation next to the boundary. A better choice is to use the squared length of the boundary edge. The same can be applied to ‘feature edges’ within a mesh. We can also formulate a constraint for points too. Volume Quadric and relation to Lindstrom Using simple vector algebra we can get the squared volume of the tetrahedron as m is the cross-prod of the edges. (4.9) is equivalent to scaling the fundamental quadric by w2/9 where w is the area of the contributing face T. Principal Components of Quadrics We can regard A as the covariance of the normals with mean 0. The eigenvectors should be roughly be in the direction of the average normal, direction of maximum curvature and direction of minimum curvature. Ambiguity of Sharp Angles
  • 3. The initial quadric assigned to both of these cases are the same since its incident segments lie along the same lines. In (b) the curve will become a single spike and quadric will allow points to move freely along the line. [Kho03] User-Guided Simplification The user can adaptively weight areas of the models. The actual weight applied is wlogV where V is the number of vertices in the input in order to get similar results of a feature from two different scales of the model. Using weights leaves more polygons in areas where user feel is important. Another tool is to use constraints – like the virtual plane in the Garland’s original work. To avoid constraints unnecessary increase cost we keep the constraint quadrics separate from the geometric ones and only add them when we are computing the optimal positions. They propose 3 types of constraints – contour, plane and point. [Lindstrom98] Fast and Memory Efficient Polygonal Simplification They pioneered the memoryless approach whereby the cost is computed wrt the current simplified meshes and not the original mesh. C(v) as a sum of squared tetrahedral volumes formed by ‘v’ and the faces around the vertex neighborhood of ‘v’ in the simplified mesh. The basic approach to finding the optimal position is to combine a number of linear equality constraints aiTv = bi, i.e. v is the intersection of three non-parallel planes in 3. If two of more of these planes are nearly parallel, minor perturbations to the plane coefficients lead to large variations in the solution. So we add a constraint to a set of existing constraints only if the plane normal does not fall within an angle of all linear combinations of the plane normals of the previous constraints. See [Lindstrom00] for a SVD based alternative. Volume Preservation By considering the signed changes in the volumes of the tetrahedrons formed by the new vertex and the old triangle. where ni is the outward normal vector of triangle ti, with magnitude twice the area of ti. Boundary Preservation
  • 4. For a planar boundary, we attempt to preserve the area enclosed by the boundary. We use signed changes in area. Volume Optimization The volume preservation constraint leaves an entire plane of candidate vertices. To further constrain the vertex position we minimize the unsigned volume of each individual tetrahedron, which is a measure of the local surface error for each triangle in the edge’s neighborhood. If all of these are coplanar the volume optimization has many solutions Boundary Optimization We minimize the sum of squared areas. Triangle Shape Optimization Which is the sum of square of the edge lengths incident upon v. This unsure the area to perimeter ration is maximized. This is used as a last resort only when both fV and fB are both close to zero. Edge Costs [Lindstrom00] Out-of-core simplification of large polygonal meshes Points are binned into grid cells, within each cell qem is used to collapse vertices and to calculate the new vertex. Only a single pass is needed through the input model. The qem used is the memoryless version from [Lindstrom98]. A hash table based on the grid cell index is used to store the input. The quadric for the triangle is computed and added to all 3 bins. Each bin represents one output vertex. Where n is the 4-vector made up of area-weighted triangle normal and scalar triple product of its vertices. Since Q is symmetric and c is not used, only 9 scalars are needed. After adding all the quadric within a grid, we use the block decomposition above and solve the Ax = b for the optimal position. They introduce a robust SVD based way to invert the quadric matrix instead of the linear constraints of [Lindstrom98].
  • 5. [Hoppe] New Quadric Metric for Simplifying Meshes with Appearance Attributes They confirmed the memoryless method of Lindstrom gives better quality. In the standard QEM scheme the Qv are computed on the original mesh and subsequently summed. As a result, the merging of the non-parallel ovals (corresponding to fine level features) gives rise to tight spherical quadrics that lock vertices and prevent further simplification, even though the resulting mesh is planar. Garland metric is quadratic in space – i.e. for m attributes it need o(n^2) space. The new quadric defines both the geometric error and attribute error based on geometric correspondence in 3D. Rather than projecting the point p onto the mesh face in an
  • 6. abstract higher-dimensional space R3+m. The error is defined as the sum of the geometric error and the attribute error. The attribute error is the squared deviation between s and the value s’ interpolated from face f at the projected point p’. The projection is done in R3. The new metric takes 11 + 4m coefficients, which is linear in m. In the memoryless scheme the Qf are computed on the face neighborhood of the collapsing edge. The memoryless simplification makes storing QEM’s unnecessary. For speed it is still good to stored the values (area(f).Qf(v)) on the faces. The new QEM sometimes shrinks the model geometry in areas of high attribute gradient – new vertex is pushed towards the center of curvature of the surface at sharp attribute transitions (see fig.7 above). Preserving volume is equivalent to a linear constraint gTvol p + dvol = 0. It is solved using a Lagrange multiplier. The volumetric gradient gvol is the sum of the face normals of Fi+1 weighted by 1/3 of their face areas. [Hoppe00] Efficient Minimizatio of New Quadric Metric for Simplifying Meshes with Appearance Attributes The minimization of the quadric metric with m appearance attribute involves solving a linear system of size (3+m)x(3+m). The system has only O(m) nonzero entries, so it can be solved in O(m2) using sparse solvers such as conjugate gradients. Here the authors show the special structure of the sparsity permits the system to be solved in O(m) time.
  • 7. [DeCoro07] Real Time Simplification using the GPU They use [Lindstrom00] qem based vertex collapse as the base and add a probabilistic octree instead of a grid. The basic alg is very simple – it use a user specified uniform grid and use the vertex shader to compute the grid index. The index is used to accumulate a per-cluster qem. In pass 2 the qem is solved using a matrix invert to get the optimal vertex position. In pass 3, the input mesh is send through the a 2 nd time, geometry shader is used to cull triangles are collapsed; otherwise the shader retrieve the positions from pass 2 and streamed out into a GPU buffer. Two warping function are introduced a) the world-view projective transform b) area-of- interest by a Gaussian. It is inverted using the Gaussian error function. The inverse warping function is used only when computing the cluster coordinates. To support adaptive resolution a probabilistic octree using hash was introduced. [Frey99] Surface Mesh Quality Evaluation Measures the aspect ratio of the set of triangles to define the quality. 1 6 Sf K Tri (M j ) Q f , where Q f . Sf is the area of the face f, pk is the half- F f F 3 p k .hk perimeter of f and hf is the longest edge. [Southern00] Evaluation of Memoryless Simplification They evaluated the common alternatives for vertex placement – a) 0 bit: mid-point b) 1 bit: half-edge collapse c) 2 bits: to encode the end-pts + mid-pt. They defined a triangle quality metric that is faster than [Frey99] 3 F e ij 1 j 1 K (M ) ( ) ; divide the edge length sum by 3 times the minimum edge. F i 1 3(min 3j 1 e ij ) F 1 For volume, they use the Gaussian Divergence Formula K vol ( M ) (v1i v2 ) v3 , i i 6 i 1