Scribd
Upload
30 views

Aam Lecture

Active appearance models are a statistical method for matching shape and appearance models to new images in an iterative process. The model is initially placed in the image, the difference or residual is measured, and the model is updated based on this residual to improve the match. The goal is to minimize the residual over iterations to find the optimal match between the model and image.

Uploaded by

Ahly Zamalek Masry
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
30 views

Aam Lecture

Active appearance models are a statistical method for matching shape and appearance models to new images in an iterative process. The model is initially placed in the image, the difference or residual is measured, and the model is updated based on this residual to improve the match. The goal is to minimize the residual over iterations to find the optimal match between the model and image.

Uploaded by

Ahly Zamalek Masry
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 29

Active Appearance Models

Suppose we have a statistical appearance


model
Trained from sets of examples
How do we use it to interpret new images?
Use an Active Appearance Model
Iterative method of matching model to
image
Interpreting Images (1)






Place model
in image
Measure
Difference
Update Model
Iterate
Quality of Match
Residual difference:
p : all parameters, eg
Ideally find and optimise p(p|r)



Cannot usually know p(r)
) ( ) ( ) ( p I p I p r
im m
=
) ( ) ( ) ( p r p r p
T
E =
) , , , , ( u s Y X
c c
c p =
) (
) ( ) | (
) | ( : rule Bayes
r
p p r
r p
p
p p
p =
Quality of Match
Usually attempt to maximise

This is equivalent to maximising

Which is equivalent to minimising
) ( ) | ( p p r p p
) ( log ) | ) ( ( log p p p r p p +
) ( log )) ( ( log ) ( p p r p p p E =
Quality of Match
Assuming independent gaussian noise:
|
|
.
|

\
|

2
2
) ( ) (
exp )) ( (
r
T
p
o
p r p r
p r
const p
r
T
+

=
2
2
) ( ) (
)) ( ( log
o
p r p r
p r
const p E
r
+ = ) ( log
2
| ) ( |
) (
2
2
p
p r
p
o
Quality of Match
If we assume all parameters equally likely
(within certain limits)



Thus we need to find the parameters which
minimise the sum of squares of residuals,

const E
r
+ =
2
2
2
| ) ( |
) (
o
p r
p
const p = ) (p
2
| ) ( | ) ( p r p = E
Quality of Match
If we assume parameters have a gaussian
distribution,


We must then minimise
) 5 . 0 exp( ) (
1
p S p p


T
p
p S p
p r
p
1
2
2
5 . 0
2
| ) ( |
) (

+ =
T
r
E
o
Optimising the Match
We must find p to minimise E(p)
p may have many (100s) of dimensions
Can put into a multi-dimensional optimiser
Likely to be rather slow
We can use some cunning approximations
to find good solution rapidly
Key Insight
Image error related to error in p
Learning the Relationship
For each of a training set
find best fit given landmarks, p
randomly perturb p by op and measure
(in model frame)
Multivariate regression to learn R in

) ( ) ( ) ( p p I p I p p r o o + = +
im m
) ( p p Rr p o o + =
More Analytic Approach
p
p
r
p r p p r
c
c
+ = + ) ( ) ( o
r r E p p
T
, minimize To = + ) E( o
T T
p
r
p
r
p
r
R
c
c
|
|
.
|

\
|
c
c
c
c
=
1
) ( p Rr p =
How Good are the Predictions?
Predicted ox vs. actual ox over test set
Multi-Resolution Predictions
Predicted ox vs. actual ox over test set
AAM Algorithm
Initial estimate I
m
(p)
Start at coarse resolution
At each resolution
Measure residual error, r(p)
predict correction op = Rr

p p - op
repeat to convergence
Search Example
Search Example
Search Example
Search Example
Sub-cortical Structures
Initial Position Converged
Sub-cortical Structures
Initial Position Converged
Brain search
Search Accuracy (Brain images)
Leave-1-out search experiments
For each image in training set:
Train model on all current image,
test on that image
average results over all images
0
10
20
30
40
50
0 1 2 3 4 5
Point-Curve Error (pixels)
F
r
e
q
u
e
n
c
y

(
%
)
Examples of Failure
Poor initialisation can lead to failure
Only samples current region
may not cover full extent of target
MR Knee Cartilage
Higher Dimensional Images
3D
2D+time
3D+time
ASM relatively straightforward
AAM problems with size of models
Problems
Automatic Model Building
Require correspondences across a set
Hard to achieve reliably
Human interaction can impart expert
knowledge
Correspondence Important
Manual Annotation Equally spaced points
Problems
Reliable measure of quality of fit
Necessary for good matching
Essential for detection (eg is object present at
all?)
RMS of residual too sensitive to positional
errors
Model initialisation
Getting good initial estimate can be hard
Problems
Failures to match at edges of model




Perhaps need some model of whole image
Multiple initialisations can help
Model explains region under itself very
well, but fails to explain all it could.

You might also like