Ivan Laptev

1
Boosted Histograms for Improved Object Detection
Ivan Laptev IRISA/INRIA, Rennes, France September
07, 2006
2
Histograms for object recognition
Remarkable success of recognition methods using
histograms of local image measurements

• Swain Ballard 1991 - Color histograms
• Schiele Crowley 1996 - Receptive field
  histograms
• Lowe 1999 - localized orientation histograms
  (SIFT)
• Schneiderman Kanade 2000 - localized
  histograms of wavelet coef.
• Leung Malik 2001 - Texton histograms
• Belongie et.al. 2002 - Shape context
• Dalal Triggs 2005 - Dense orientation
  histograms

Likely explanation Histograms are robust to
image variations such as limited geometric
transformations and object class variability.
3
Histograms What vs. Where
What to measure?
Histograms
Where to measure?

• No guarantee for optimal recognition
• Different regions may have different
  discriminative power

4
Idea
selected features
boosting
weak classifier
? ? ?

• Efficient discriminative classifier
  FreundSchapire97
• Good performance for face detection
  ViolaJones01

AdaBoost
Haar features
SVM Neural Networks
Histogram features
Too heavy
5
Weak learner
Possible approach
1-dim. projections onto predefined vectors
Example 1
6
Weak learner
Possible approach
1-dim. projections onto predefined vectors
Example 2
7
Fischer weak learner
Alternative approach

• Assume Normal distribution of features
  (hopefully valid at least for some of 105
  features!)
• Compute projection direction by FLD

• Can be modified to minimize the error of
  weighted samples (required for boosting)

feature mean feature covariance
Fischer learner
1-bin learner
Evidence from real image training data
8
Histogram features
9
Training data
10
Training Selected Features
0.999 correct classification 10-5 false positives
376 of 105 features selected
11
Object detection

• Scan and classify image windows at different
  positions and scales

• Cluster detections in the space-scale space
• Assign cluster size to the detection confidence

12
PASCAL Visual Object Classes Challenge 2005
(VOC05)
217 / 220
motorbikes
bicycles
123 / 123
people
152 / 149
cars
320 / 341
13
Evaluation criteria
Ground truth annotation

• Detection results
• gt50 overlap of bounding box with GT
• one bounding box for each object
• confidence value for each detection

Precision-Recall (PR) curve
Average Precision (AP) value
14
Evaluation of detection
PR-curves for the Motorbike validation dataset
FLD learner
1-bin classifier
15
Results for VOC05 Challenge
People test1
Bicycles test1
cars test1
Motorbikes test1
16
Results for VOC05 Challenge
Average Precision values
17
(No Transcript)
18
(No Transcript)
19
PASCAL Visual Object Classes Challenge 2006
(VOC06)
20
Results for VOC06 Challenge
Competition "comp3" (train on VOC data) Class
bicycle"
examples
21
Results for VOC06 Challenge
Competition "comp3" (train on VOC data) Class
cow"
examples
22
Results for VOC06 Challenge
Competition "comp3" (train on VOC data) Class
horse"
examples
23
Results for VOC06 Challenge
Competition "comp3" (train on VOC data) Class
motorbike"
24
Results for VOC06 Challenge
Competition "comp3" (train on VOC data) Class
person"
25
Results for VOC06 Challenge
Average Precision values
26
Final Notes

• All results are obtained with a single set of
  parameters
• Small number of training samples is sufficient
• Efficient detection 10fps on 320x280 images
• Extension to texton/color histogram features is
  straightforward

Open questions

• Other free-shape regions better? How to find
  them?
• Better weak learner that takes advantage of
  histogram properties
• View transformations

27
Final Notes

• All results are obtained with a single set of
  parameters
• Small number of training samples is sufficient
• Efficient detection 10fps on 320x280 images
• Extension to texton/color histogram features is
  straightforward

Open questions

• Other free-shape regions better? How to find
  them?
• Better weak learner that takes advantage of
  histogram properties
• View transformations

28
Final Notes

• All results are obtained with a single set of
  parameters
• Small number of training samples is sufficient
• Efficient detection 10fps on 320x280 images
• Extension to texton/color histogram features is
  straightforward

Open questions

• Other free-shape regions better? How to find
  them?
• Better weak learner that takes advantage of
  histogram properties
• View transformations

29
Final Notes

• All results are obtained with a single set of
  parameters
• Small number of training samples is sufficient
• Efficient detection 10fps on 320x280 images
• Extension to texton/color histogram features is
  straightforward

Open questions

• Other free-shape regions better? How to find
  them?
• Better weak learner that takes advantage of
  histogram properties
• View transformations

30
Final Notes

• All results are obtained with a single set of
  parameters
• Small number of training samples is sufficient
• Efficient detection 10fps on 320x280 images
• Extension to texton/color histogram features is
  straightforward

Open questions

• Other free-shape regions better? How to find
  them?
• Better weak learner that takes advantage of
  histogram properties
• View transformations

• Detection tasks in VOC05,VOC06 are far from
  being solved, it is a challenge!

31
(No Transcript)