Deformable Template as Active Basis

1
Deformable Template as Active Basis Zhangzhang
Si UCLA Department of Statistics Ying Nian Wu,
Zhangzhang Si, Chuck Fleming, Song-Chun Zhu
ICCV07 (the work in this talk is outdated, see
http//www.stat.ucla.edu/ywu/AB/ActiveBasisMarkI
I.html for the updated results in our IJCV paper)
2
Motivation
Design a deformable template to model a set of
images of a certain object category. The template
can be learned from example images.
2009-11-10
CIVS, Statistics Dept. UCLA
2
3
Related work

• Representation generative and deformable models
• Sparse coding Olshausen-Field 96
• Deformable templates Yuille-Hallinan-Cohen 89
• Active contours Kass-Witkin-Terzopoulos 87
• Active appearance Cootes-Edwards-Taylor 95
• Texton model Zhu et.al. 02
• Computation learning and pursuit algorithm
• 1. Matching pursuit Mallat and Zhang 93
• 2. HMAX Riesenhuber-Poggio 99, Mutch-Lowe 06
  
• 3. Adaboost Freund-Shapire 96, Viola-Jones
  99

4
Linear additive image model
Image reconstruction by matching pursuit.
selected from a dictionary of Gabor wavelet
elements
location
scale
orientation

• Two extensions
• Encoding a single image
  Simultaneously encoding a set of images
• Allow each Gabor wavelet element Bi to locally
  perturb.

5
The active basis model
(Gabor elements represented by bar)
Active Local perturbation
When encoding image Im, we use the perturbed
version of Bi
6
Deformable template using active basis
A car template
(Gabor elements represented by bar)
2009-11-10
CIVS, Statistics Dept. UCLA
6
7
Deformable template using active basis
A car template
8
Learning the template pursuing the active basis
q(I) background distribution (all natural
images) p(I) pursued model to approximate
the true distribution.
Example images
Gabor elements selected
9
Pursuing the active basis
MLE
(Projected on B1,,Bn)
(orthogonality of B1,,Bn)
2009-11-10
CIVS, Statistics Dept. UCLA
9
10
Pursuing the active basis
2009-11-10
CIVS, Statistics Dept. UCLA
10
11
Shared pursuit algorithm
2009-11-10
CIVS, Statistics Dept. UCLA
11
12
Learning the template pursuing the active basis
A car template consisting of 60 Gabor elements
Car instances
13
Experiment 1 learning an active basis model of
vehicle
template

• 37 training images, listed in the descending
  order of log-likelihood ratio
• 4.3 seconds (Core 2 Duo 2.4GHz) , after
  convolution

14
Experiment 2 learning without alignment
Active basis pursuit EM
Given bounding box for the first example for
initialization. Iterate - Estimate the
bounding boxes using current model. -
Re-learn the model from estimated bounding boxes.
15
Experiment 3 learning and clustering
16
Experiment 4 car detection with active basis
model

• Scan bounding box over the image at
  multi-resolutions
• Compute log-likelihood ratio by combining
  responses from active basis

LLR log likelihood ratio
LLR log likelihood ratio
map of LLR at optimal scale
Maximum LLR over scale
17
Experiment 5 head-and-shoulder recognition
Features using the same set of Gabor filters.
Some negatives
Some positives
Negatives include various in-door and out door
scenes, with and without human
Human head and shoulders, roughly aligned
43 training positives, 157 training negatives 88
testing positives, 474 testing negatives
18
Experiment 5 head-and-shoulder recognition
comparing with Adaboost
ROC of sigmoid model is a further improvement of
the result presented in the paper.
19
Main contributions
1. An active basis model as deformable
template. 2. An active bases pursuit algorithm
for fast learning.
http//www.stat.ucla.edu/ywu/ActiveBasis.html Do
wnload 1) Training and testing images 2) Matlab
and mex-C source codes that reproduce all the
experiments in the paper and powepoint.