Scene Classification

1
Scene Classification

• Pulkit Agrawal
• Y7322
• BVV Sri Raj Dutt
• Y7110
• Sushobhan Nayak
• Y7460

2
Outline

• What is a scene
• Scene recognition
• Method
• Results
• Future Work
• References

3
What is a Scene?

• Scene- as opposed to object or texture
• Object when view subtends 1 to 2 meters around
  observer---hand distance

4
What is a Scene?

• observer and fixated point- gt5 meters

5
Scene Recognition

• 2 approaches
• Object recognition
• Global info details and object info ignored
• Experimental evidence
• Gist of image

6
Scene Recognition

• Exclusive classification

• Structural attributes- Continuous organization of
  scenes along semantic axes

7
Semantic axes

• 2 levels
• Degree of naturalness man-made to natural
  landscape
• Ambiguous (building in field) pictures around
  center

8
Semantic axes

• Natural scenes- degree of openness
• Artificial urban scenes- degree of verticalness
  and horizontalness
• Highways--?Highways Tall Building---?Tall
  Buildings

9
Method
Information at various Scales
What do we Need ??
High Frequency ?
Low Frequency ?
Both ??
10
Feature Extraction
Image Power Spectrum
Gabor Filters (Scale, Orientation)
Features (512 used)
11
Mathematical Details

• Important data from Image power spectrum
• Structural discriminant feature
• DSTDiscriminat Spectral Template- --an encoding
  of the discriminant structure between two image
  categories
• u -?weighted integral of power spectrum

12
Classification
Required Classes
Image Feature Vector()
Linear Discriminant Analysis
Discriminating Vector (D.V)
Maximum Separation b/w classes
13
Mathematical Details..

• Image represented as Feature Vector x.
• m1 , m2 mean vector of feature vector of 2
  classes

14
Mathematical Details

• gn feature
• Gn Gabor filter
• dn through learning

15
Learning
Projection of Training Set Image F.V. on D.V.
Use LDA to determine Threshold
Classifier Obtained
16
Learning
17
Work..
Artificial v/s Natural
Open v/s Non Open
18
Results
Artificial v/s Natural

• Natural
• 80 Test Images
• 75 classified Correctly

• Artificial
• 80 Test Images
• 67 classified Correctly

89 Correct results
19
Result
20
Future Work

• Arrangement in semantic axes
• Addition of features
• Depth Symmetry
• Contrast
• Ruggedness
• 8 category arrangement (skyscrapers, highway,
  street, flat building, beach, field, mountain,
  forest)
• Experiment with Haar and other filters

21
Reference

• Torralba A. Olivia A., Semantic Organisation of
  Scenes using Discriminant Structural Templates
  (1999)
• Torralba A. Olivia A., Modeling the Shape of
  the Scene A Holistic Representation of the
  Spatial Envelope(2001)
• Olivia A., Gist of the Scene
• http//people.csail.mit.edu/torralba/code/spatiale
  nvelope/