study latent doodle space
DESCRIPTION
studyTRANSCRIPT
Latent Doodle SpaceLatent Doodle Space
William Baxter1, Ken-ichi Anjyo2
OLM Digital, Inc.EUROGRAPHICS,2006
Presented by C.M. Hsu
OLM Digital, Inc.
EUROGRAPHICS,2006
AbstractAbstract
AbstractAbstract
Major tech. A heuristic algorithm to match strokes
between the inputs. Extract a low dimensional latent doodle
space from the inputs.
ApplicationsApplicationsThe Randomized Stamp ToolThe Randomized Stamp Tool
InputInput
OutputOutput
ApplicationsApplicationsHandwriting synthesisHandwriting synthesis
OutputOutput
InputInput
OutlineOutline
Overview Related Work Stroke Matching Algorithm Building the Latent Doodle Space Conclusions and Future Work Demo
OverviewOverview
Overview
1. Assignments of stroke correspondence
2. Resample corresponding strokes with the samesame number
of sample points
3. Reverse the parameterization of strokes to improve the point-to-point
correspondence
4. Build the latent control space
1. Similar to Computer-assisted in-between algorithms2. N-way correspondences, not pair3. Competitive-learning algorithm
•K-means-like•Match stokes based on Kuhn-Munkres method•O(N3)
Synthesis
PCA
thin plate spline RBF
PCA
GP
Feature vector
GPLVM
Related WorkRelated Work
Related Work – Stroke Correspondences
The order of strokes between two images are identical, [Burtnyk N., SIG75]
Closed shape only, [Sederberg, SIG92]
Related Work – A low-dimensional latent space
Eigenfaces by PCAPCA, [Turk, Nerosci91] Multidimensional motion interpolation b
y Radial basis functions (RBFRBF) [Rose et al.,IEEE98]
The Gaussian Process Latent Variable Model (GPLVMGPLVM), [Lawrence, NIPS04]
Create keyframe from ex. by GPLVMGPLVM, [Grochow K., CV04]
Related Work Create many drawings from a few ex., [Ko
var, UIST01]
Stroke Matching Stroke Matching AlgorithmAlgorithm
1. Finding Stroke Correspondences2. The Assignment Cost Matrix
3. Stroke Re-sampling and Alignment
Finding Stroke Correspondences
K-means like1. Initialize stroke-to-cluster assignment
Clustering by the drawn order of strokes simply
2. Update the cost matrix3. Reassign stroke based on new clusters
Linear assignment problem (strokes clusters) Constrain: one stroke per drawing to each cluster Kuhn-Munkres algorithm, (N3), N as number of strokes
4. If reassignments made, goto 2
The Assignment Cost Matrix E=ed+ec+et
Translation error, ed The mean of the stroke differs from the mean of t
he cluster Orientation and eccentricity error,ec
The covariance axe of the stroke differs from those of the cluster
Topological matching cost, et The connectivity of the stroke differs from the con
nectivity of the members of the cluster
Translation error
ed: The mean of a stroke differs form the mean of the cluster
2csd -e
stroke theofpoint mean the:s cluter theofpoint mean the:c
Orientation and eccentricity error
ec: The covariance axe of the stroke differing from those of the cluster
f2cs
cc A-Awe
Stroke 1Stroke 2
2212
2111
ssss
sssssC
c2
c1
s 2,
s1
cs
cs
cs
1 1j
2
ijf2
f2
c
4s2
s1
s2
s1c
clusters. and strokes of eeigen valu the:],[
eeigen valuby ordering
C, of orseigen vect ofmatrix the:],[
clusters and strokes ofmatrix covarince the:]C,[C
b norm, Frobenius :
16w1,,/avg,maxw
AA
Bm
i
n
Topological matching cost et: Connectivity cost of the stroke differing from
the connectivity of the members of the cluster
CCssjj: the number of strokes the ith stroke in the same drawing.
CCccjj : the number of clusters the ith cluster by average distance.
2ci
sjt c-c0.2e
S2S1
S3
S2 is connected to S1, cs2=1 C2 is connected to C1, cc
2=1
C2
C1
Stroke Re-sampling and Alignment Same number of points on the correspon
ding strokes for RBF, Gaussian process regression
Reverse backward strokes The total distance error between two stroke
s is lower when the point ordering is reversed
Building the Latent Doodle Space
PCA2 principal
components
thin plate spline RBF
PCA2 principal
components
GP
Feature Vectorm = {p1,p2,..pn} of Strokem
GPLVM
RBF•Gaussian•Thin plate spline, r4logr
bestinput output
Conclusions and Future Work Using machine learning
Find good assignment weights in cost functionEx: S.T. like a support-vector classifier could be trainned to assign strokes to clusters.
Allow user to appraise the products form as latent space. [Kovar, UIST01]
Accept scanned drawing Accept completely free-form hand-drawn sketc
h without the line constrain of uniform width.
EndEnd