A 4-WEEK PROJECT IN Active Shape and Appearance Models - PowerPoint PPT Presentation

1 / 23
About This Presentation
Title:

A 4-WEEK PROJECT IN Active Shape and Appearance Models

Description:

A 4-WEEK PROJECT IN Active Shape and Appearance Models by Ren e Anderson and Vedrana Andersen Overview Preprocessing (Ren e) Building models (Vedrana) Manipulating ... – PowerPoint PPT presentation

Number of Views:145
Avg rating:3.0/5.0
Slides: 24
Provided by: VedranaA7
Category:

less

Transcript and Presenter's Notes

Title: A 4-WEEK PROJECT IN Active Shape and Appearance Models


1
A 4-WEEK PROJECT INActive Shape and Appearance
Models
  • by Renée Anderson
  • and Vedrana Andersen

2
Overview
  • Preprocessing (Renée)
  • Building models (Vedrana)
  • Manipulating faces (Renée)
  • Active appearance models (Vedrana)

3
STATISTICAL MODELS Concept
  • Several steps developed by Cootes and Taylor
  • Data capture and representation
  • Data normalization
  • Statistical analysis

4
PREPROCESSINGData
For each object (face, heart, spine)
  • Image
  • Landmark points

5
PREPROCESSINGCropping and Translation
  • Cropping out the background
  • Center of gravity of landmark points in the
    center of image

6
PREPROCESSINGAlignment
  • Allowed transformations translation, rotation
    and scaling
  • Iterative method aligning each shape with mean
    shape

7
PREPROCESSINGMasking
  • For visualization
  • Leave inside of convex hull of landmark points

8
PREPROCESSINGWarping
  • Mapping landmark points to mean shape
  • Tried two- piecewise linear- thin-plate spline

9
PREPROCESSINGGray-level Normalization
  • Applying scaling and offset to gray-level vectors
    (samples)
  • Iterative method normalizing each gray-level
    vector with mean gray-level

10
BUILDING THE MODELSNormalized Data
  • Each object contributes with
  • Landmark points, aligned
  • Shape-free patch, normalized
  • Statistical analysis (PCA)

11
BUILDING THE MODELSPrincipal Component Analysis
  • Objects (shapes, gray-levels) as vectors
  • Distribution of s points in n- dimensional space
  • Alternative coordinate system- no
    correlations- sorted coordinates

12
BUILDING THE MODELSPrincipal Component Analysis
  • Method finding eigenvectors and eigenvalues of
    the covariance matrix
  • Linear model

13
BUILDING THE MODELSShape Model
  • Shape vector
  • Model

14
BUILDING THE MODELSGray-level Model
  • Gray-level vector samples of gray values
  • Model

15
BUILDING THE MODELSAppearance Model
  • Appearance vector
  • Model

16
BUILDING THE MODELSAll Faces Modes of Variation
First mode
Third mode
Second mode
17
BUILDING THE MODELSNeutral - Appearance Variation
  • Changing the first three modes of variation
    simultaneously

18
BUILDING THE MODELSHearts - Appearance Variation
  • Changing the first three modes of variation
    simultaneously

19
MANIPULATING FACESReconstruction
20
MANIPULATING FACESCaricatures
21
MANIPULATING FACESForced Smiles
22
ACTIVE APPEARANCE MODELSConcept
  • Search optimization to minimize
  • Learning estimating A
  • Iterative model refinement

23
Thank you!
Write a Comment
User Comments (0)
About PowerShow.com