FEASIBILITY OF USING WAVELET BASED PYRAMIDAL ANALYSIS FOR VISUAL CONTENT IMAGE DESCRIPTION

1
FEASIBILITY OF USING WAVELET BASED PYRAMIDAL
ANALYSIS FOR VISUAL CONTENT IMAGE DESCRIPTION

• Manuel Agustí, J. M. Valiente
• Computer Engineering Department (DISCA)
• Polytechnic University of Valencia (UPV), Spain
• magusti, jvalient_at_disca.upv.es
• 3rd IASTED International Conference on
  Visualitzation, Imaging and Image Processing
• (VIIP 2003)
• Benalmádena, Málaga, September 2003

2
Index

• Visual Content Image Description
• Image Retrieval
• CBIR
• What an Image is?
• Image Database
• Image Modeling
• Wavelet, Multiresolution scenario
• The Piramidal approximation
• Borders, Walking throught the pyramid, Spatial
  Accurate?
• Examples of application

3
Visual Content Image Description

• Multimedia databases images.
• Terms
• Vagueness.
• Indexing (ordering/ranking).
• Di/Similarity (exactly has no meaning in this
  context).

• The significance of the visual elements depends
  on the subjective of the observer, and it is not
  know before to create the image database. So, it
  is necessary to develop search criteria and they
  must be able to manage vagueness and incomplete
  descriptions from the users.

4
Image Retrieval

• Previous work structural analysis
• Based on objets, groups, and composition
• Smooth, Segmentation Labelling
  Vectorization
• Comparision Grouping Reconstruction
  
  Fundamental Paralelogram
• Theory of Simetry Groups, point simetry, ...
• Reconstruction

5
Image Retrieval

• Content Based Recovery Model
• Space of representation of images image domain.
• CBIR is build on concepts
• Image Modelling
• Relevance content for the user (e. g. Relevance
  Feedback).
• Search Techniques
• Natural way to express features and to use vague
  information.
• Similarity Metrics
• For human perception.
• Image Modeling guides the other two

6
Image Retrieval CBIR

• Focus on features?
• CANDID, Chabot, WebSeek, QBIC, PhotoBook.

• Unified framework --gt Multiresolution Analysis

7
What an Image is?
An easy game What is this image about?
8
What an Image is?
An easy game What is this image about?
9
What an Image is?
An easy game What is this image about?
10
What an Image is?
An easy game What is this image about?
11
What an Image is?
An easy game What is this image about?
12
What an Image is?
An easy game What is this image about?
13
What an Image is? Image Data Base

• Examples of images in our context

14
Image Modelling

• Multiresolution using wavelet decomposition.

• Scenario
• Colour images (normalised in size).

15
Image Modelling

• Multiresolution using wavelet decomposition.

• Scenario
• Colour images (normalised in size).
• Pyramidal transformation (9 levels).
• Haar, Daubechies, Burt-Adelson, Battle-Lemarie,
  Symmlet, B-Spline, and Coiflets.

16
IM Multiresolution scenario
Level 1
.. Level N
Level 2
2
1
A
D1
1
1
D1
A
1
1
D3
D2
1
D2
1
D3
Ai f(Ai-1, D1i-1, D2i-1, D3i-1)
17
The pyramidal perspective

• Multiresolution analysis from a pyramidal
  sequence
• Retains the important things (resembles the
  subjective point of human observer).
• Complete transformation (top/down bottom/up).
• Separation information persistent and detail.

18
Border extraction

• Visual correspondence with detail and persistent
  information that is behind colour.
• border(I) level(0, I) - level(1, I)

19
Walking through the pyramid

• In the reverse order
• Considering the incorporation of de detail
  information at each level

20
Spatial accurate?

• Pixel position of the reconstructed vs the
  original image.

• Extract constitutive elements of an image

21
Examples of application

• Extract constitutive elements of an image
• Tolerance to ilumintation conditions.
• Extract objects.
• Can they be described in a perceptually way?
• There are elements that are more perceptually
  important in an object
• How are related this objects?
• Are they forming groups?
• At what level?

22
Borders
23
Borders
Level 0
Level 1
Level 8
24
Borders
25
Map
26
Map
Level 0
Level 1
Level 8
27
Map
28
Conclusions

• Image Modelling using wavelets
• Extract constitutive elements of an image to
• Guide a segmentation process.
• Guide a clustering process.
• Extract the minimal area or the fundamental part.
• Extract information in terms of spatial
  distribution.
• Describe image content in terms of features at an
  abstract level.
• Search Techniques
• Improve lt-- Early detection of differences.
• Similarity Metrics
• Capture spatial information visual features

29
Thank you for your attention! Any question?
30
FEASIBILITY OF USING WAVELET BASED PYRAMIDAL
ANALYSIS FOR VISUAL CONTENT IMAGE DESCRIPTION

• M. Agustí, J. M. Valiente
• Computer Engineering Department (DISCA)
• Polytechnic University of Valencia (UPV), Spain
• magusti, jvalient_at_disca.upv.es
• 3rd IASTED International Conference on
  Visualitzation, Imaging and Image Processing
• (VIIP 2003)
• Benalmádena, Málaga, September 2003