Progress Report

April 26th, 2005

Present Snapshot

• Possible collaboration with Bill Crum
• Normalisation in model evaluation
• Locally-orderless images (Lewis Griffin)
• Possibility of a journal paper
• ECCV, ISBI deadlines - end of 2005

Weekend Activities

• Files (code, documentation) and diary organisation
• Result of a hectic month of work

Evaluation Methods - Comparison

• Exchange of dataset (UCL) was suggested
• Comprison will be simpler with identical data
• Leading to better understanding of the algorithms
• Quantitative comparison is hard
• Qualitative (or relative) comparison -- easier

Overlap Measures for Validation

• Requires ground truth
• 36 2-D brains with ground truth in ISBE
• 8+ 3-D brains with ground truth from CMU
• Overlap-based method might be weak
• Pair-wise registration is shown to surpass group-wise

Comparison 'Protocol'

• Need to descend to 2-D
• Decide on dataset to use
• Decide on registration methods
• Standardise registration output type
• Evaluate based on the available measures
• Compare graphs

Locally-orderless images

• The E-mail cites relevant literature
• Room for investigation (versus shuffle distance)
• Histogram-based approach involving 'strexture'

Locally-orderless images

• Griffin LD (1995) Scale-imprecision space. IVC. 15(5):369-398
• Koenderink JJ & van Doorn AJ (1999) The structure of locally-orderless images. IJCV 31(2-3):159-168

Locally orderless images

• Van Ginneken B & Romeny BMT (2000) Applications of locally orderless images. J Vis Comm Image Rep. 11(2):196-208
• Florack L & Duits R (2003) Regularity classes for locally orderless images. LNCS vol. 2695 255-265

Normalisation

• Normalisation framework in the past

Summary

• The current plan can be broken apart into:
  • Further experiments
  • Improved evaluation method
  • Comparing evaluation methods
  • Wrapping up methods more coherently

Next Steps

• Investigation of models
• Large experiments which involve different data
• Locally-orderless images
• A normalisation framework is desirable