MedIX

1
MedIX Summer 05

• Wavelet-based texture analysis and segmentation

2
Done so far

• Given a pre-segmented organ region, can you tell
  me what it is kidney, heart etc?
• It depends on its texture
• Texture features extracted from the wavelet
  transform of the image
• Wavelet-based texture classification of organ
  tissues from CT- scans

3
Wavelet transform (an example)
Averages Horizontal Activity
Vertical Activity Diagonal Activity
4
Haar Wavelet
A D


A D
Original image
9 7 3 5
A
D
A
D
8 4
1 -1
Averages
Details
6
2
1 -1
6 2 1 -1
Wavelet coefficients
AA AD
DA DD
5
Texture Classification Process
Improving the
Organ/Tissue Image

• Other Wavelets
• Wavelet frames
• Gabor filters

• Other descriptors

• Different classifier
• Similarity measures

• Different criteria for picking the most
  discriminatory features

6
Texture segmentation

• Given an image, can you tell me which/how many
  organs you have?
• Identifying regions with similar texture

7
Wavelet-based texture segmentation

• Block-wise segmentation using feature extracted
  from the wavelet transform
• Which distance b/w blocks?
• Which features?
• What threshold?
• Split and merge techniques
• Pixel-level texture segmentation

8
On the fence

• Ridgelets a variation of the wavelet transform
  particularly efficient in detecting line-like
  edges/phenomena
• Curvelets a variation of wavelets/ridgelets
  particularly efficient in detecting curve-like
  edges/phenomena
• Application to segmentation