Lyndsey Clarke

1
Elastic Facial Caricature Warping

• Lyndsey Clarke

PhD Student University of Wales Swansea
http//cs.swan.ac.uk/cslyndsey
2
Introduction

• Elastic Facial Caricature Warping is a computer
  graphics technique that I have developed during
  my first year as a PhD student.
• The project began life as my undergraduate
  dissertation.
• A short paper on this technique has been
  published in Eurographics 2006 Short Papers
  proceedings, and a poster was presented at the
  conference last month.
• The two main themes that have been developed are
• Caricaturisation
• Beautification
• This talk aims to explain the paper in a little
  more detail and give you all an insight into my
  research.

http//cs.swan.ac.uk/cslyndsey
3
What exactly is Elastic Facial Caricature
Warping?
The Subject
2. Beautification
1. Caricaturisation
3. Deviation
Hes Beautiful!
Weve gone too far!
Hee! Hee!
Ha! Ha!
Uh oh
4
The History of the Caricature

• Caricatures are portraits that exaggerate or
  distort the essence of a person to create an
  easily identifiable, but humorous, visual
  likeness.
• Some of the earliest caricatures are found in the
  works of Leonardo da Vinci, who actively sought
  people with deformities to use as models!
• Caricatures gained further popularity in the
  early 19th century, when satirical drawings of
  politicians and local celebrities would be
  printed in newspapers, however they were often
  not warmly received.
• After WW1 the appeal of caricatures rose
  dramatically and they became a very common sight
  in most magazines and newspapers due to large
  demand from the public.

Early Caricature of Charles Darwin (1880)
Modern Caricature of Arnold Schwarzenegger
http//cs.swan.ac.uk/cslyndsey
5
Caricaturisation

• Caricaturisation is an Art
• Caricatures are a form of visual art, in which
  subjects are portrayed with exaggerated features
  that emphasise their individual differences from
  everyone else.
• L. Redman, How to Draw Caricatures, Contemporary
  Books, 1984
• The aim of a caricature is to provide an amusing
  source of entertainment and humour, especially in
  western society.
• Caricaturisation is a Science
• A caricature is typically drawn by emphasising
  all of the features that make that person
  different from a completely symmetrical reference
  face, perceived by the artist to be the norm.
• Craniofacial Anthropometry can be used to provide
  important facial measurements that should be
  considered during the creation of the caricature.
  An example of some of these measurements are
  shown on our reference image on the right.
• Caricaturisation is Computer Graphics
• Our aim therefore is to unite the disciplines of
  art and science by introducing a novel and
  entertaining, physics based approach to the
  creation of caricatures using computer graphics.

A Caricature
Our Reference Face with Feature Points
http//cs.swan.ac.uk/cslyndsey
6
Beautification

• Beautification is the process of physically
  making something appear more beautiful by
  altering its physical structure in some way.
• Facial beauty is characterised by the following
  factors
• SYMMETRY
• PROPORTION OF FEATURES
• FEATURE PLACEMENT
• PROFILE OUTLINE
• BABYFACEDNESS
• According to some research, the more perfect a
  persons face is, the healthier they should be.
• Like Caricaturisation, Beautification is also
  related to average face construction, since a
  perfectly average face is considered to be more
  beautiful than a standard face such as mine or
  yours, because all the factors I just metioned
  are obeyed.

Brachycephalic
Dolichocephalic
Straight
http//cs.swan.ac.uk/cslyndsey
7
Some Symmetrical Faces
Symmetry of left side of face
Symmetry of right side of face
Subject
8
The System Pipeline
1. Input
3. Calculation
2. User Interaction
4. Result
Feature Specification
System computes displacements and applies
chosen model to image
User chooses model
Scheme A Scheme B Scheme C Scheme D
Source Image
http//cs.swan.ac.uk/cslyndsey
9
Feature Specification

• Feature Specification
• How do we calculate the necessary displacements?
• The distance between each pair of corresponding
  feature points on the reference and subjects
  face encodes a measurement used in the
  transformation.
• These measurements and then used in our physical
  model to create the caricature.
• More formally
• Let Iin be an input image of a subject to be
  caricaturised Iref be a reference image,
  representing the norm and Iout be the output
  image of the caricature to be generated. Given a
  feature point p on image Iin and its
  corresponding feature point r on Iref, we would
  like to determine q on Iout as the resulting
  exaggerated feature point

http//cs.swan.ac.uk/cslyndsey
10
The Force - Motion Model

• This is the first model that was developed, and
  it works by imagining that each feature point on
  the subject is a particle in motion.
• Mathematically it is based on Newtons Laws of
  Motion, given as
• Two different schemes were developed in this
  model
• Scheme A Controlling by Time
• Each particle moves from its origin over a
  user-defined period of time.
• Scheme B Controlling by Force Mass
• As an extension to A, each particle can be
  assigned a weight and a global or local force can
  also be applied.

http//cs.swan.ac.uk/cslyndsey
11
The Pseudo Stress Strain Model

• It is natural to associate caricaturisation with
  the process of deformation by stretching or
  compressing different parts of a face. This
  directly leads to the use of metaphors related to
  material properties and mechanics.
• Our pseudo stress-strain model can be given as
• Users can apply different virtual material
  values to the facial features in order to produce
  different caricatures.
• We have developed two separate schemes in this
  model
• Scheme C Non-Uniform Original Length
• Scheme D Uniform Original Length

http//cs.swan.ac.uk/cslyndsey
12
Virtual Material Properties

• Instead of explicitly specifying displacements,
  users can assign virtual physical and material
  properties to the face in an intuitive manner.
• Not only does this provide users with intuitive
  metaphors for specifying a desired
  caricaturisation, but it also introduces a novel,
  and more entertaining aspect to the caricature
  generation process.

Virtual Material Hookean coefficient Non-Hookean coefficient Maximum Strength
Rubber 0.01 0.1 7
Polystyrene 2.8 0.9 35
Wood 12 0.95 75
Brick 18 1.2 39
Glass 65 2 500
Gold 83 0.92 120
Steel 205 0.99 1000
Diamond 950 1 12605
Some Material Properties used by the System
http//cs.swan.ac.uk/cslyndsey
13
An Example Scheme D Uniform Original Length
Concrete
Polystyrene
Rubber
Wood
Silicone

• An assumption is made that the line segment
  between reference and subject points has been
  stretched from its original length (u). This
  value is then used with any two material value
  properties to complete the warp.

http//cs.swan.ac.uk/cslyndsey
14
Future Work Possible Applications

• There are many areas that I will be exploring in
  order to extend this work further for my PhD.
• Currently however the main theme I am working on
  is extending the system so that it produces a 3D
  result from 2D input images using depth maps.
• Possible Future Applications
• Edutainment Software for school children.
• Plastic Surgery prediction aid.
• General entertainment/fun software for use by
  everyday people in their own homes.

3D caricature of Albert Einstein
15
Some Results
http//cs.swan.ac.uk/cslyndsey
16
THANK YOU! Any Questions?
http//cs.swan.ac.uk/cslyndsey