Content-Based Image Retrieval: Feature Extraction Algorithms EE-381K-14: Multi-Dimensional Digital Signal Processing

1
Content-Based Image RetrievalFeature Extraction
AlgorithmsEE-381K-14 Multi-Dimensional Digital
Signal Processing

• BY Michele Saad
• EMAIL michele.saad_at_mail.utexas.edu
• PROF Brian L. Evans

2
Motivation

• Increased use of image and video
• Education
• Entertainment
• Commercial purpose
• Need for efficient and effective browsing into
  image databases
• Need for reduction of semantic gap between
  low-level features and high-level user semantics

3
Objectives and Contributions

• Objective
• Implementation and comparison of texture and
  color feature extraction algorithms
• Contribution
• An up-to-date comparison of state-of-the-art
  texture and color feature extraction methods

4

• 1

5
Color Features
Color Feature Pros Cons Color Space
Conventional Color histogram Fast computation Simple High dimensionality No color similarity No spatial info HSV
Fuzzy Color Histogram Fast Computation Color similarity Robust to quantization noise Robust to contrast High dimensionality More computation Appropriate choice of membership weights needed HSV
J. Huang, S. R. Kumar, M. Mitra, W. J. Zhu and R.
Zabih, Time Indexing Using Color Correlograms,
Proc. IEEE Conf. on Computer Vision and Pattern
Recognition, pp. 762 768, June 1997
6
Color Features Contd
Color Feature Pros Cons Color Space
Correlogram Spatial Info Very slow High dimensionality No color similarity HSV
Color/Shape Method Spatial info Area Shape More computation Sensitive to clutter Choice of appropriate color quantization thresholds needed HSV
N. R. Howe, D. P. Huttenlocher, Integrating
Color, Texture and Geometry for Image Retrieval,
Proc. IEEE Conf. on Computer Vision and Pattern
Recognition, vol. II, pp. 239-246, June 2000.
7
Color Image DatabaseThe Corel Database

• 10 classes of 100 images each

• http//wang.ist.psu.edu/IMAGE

8
Color Feature ExtractionRetrieval Results
CCH FCH Correlogram Color/Shape
Average Retrieval Score 80.12 82.05 69.48 70.03
NB Euclidean distance measure used
9

• 2

10
Texture Features
Texture Feature Pros Cons Frequency Domain Partition
Steerable Pyramid Supports any number of orientation Sub-bands undecimated
Contourlet Transform Lower sub-bands decimated Number of orientations is a power of 2
S. Oraintara, T. T. Nguyen, Using Phase and
Magnitude Information of the Complex directional
Filter Bank for Texture Image Retrieval, Proc.
IEEE Int. Conf. on Image Processing, vol. 4, pp.
61-64, Oct. 2007
11
Texture Features Contd
Texture Feature Pros Cons Frequency Domain Partition
Gabor Wavelet Highest retrieval results Over-complete representation Computationally intensive
Complex Directional Filter Bank Competitive retrieval results More computation
S. Oraintara, T. T. Nguyen, Using Phase and
Magnitude Information of the Complex directional
Filter Bank for Texture Image Retrieval, Proc.
IEEE Int. Conf. on Image Processing, vol. 4, pp.
61-64, Oct. 2007
12
Texture Database The Brodatz Database

• 13 different textures
• Bark, brick, bubbles, grass, leather, pigskin,
  raffia, sand, straw, water, weave, wood and wool
• Rotated at different angles
• Examples

http//www.ux.uis.no/tranden/brodatz.html
13
Texture Feature ExtractionRetrieval Results
Steerable Pyramid Contourlet Transform Gabor Complex Directional Filter Bank
Average Retrieval Score 63.02 63.67 81.48 76
NB L1 Norm used in the distance measure
14
Conclusion and Future Work

• Highest retrieval results obtained by
• Fuzzy color histogram
• Gabor wavelet transform
• Keeping in mind some trade offs
• Appropriate distance measures need to be
  considered further
• May improve results further