Progress Report

May 9th, 2006

Overview

Experiments Proposed in the Last Meeting

Experiments Proposed in the Last Meeting - Ctd.

Experiments Proposed in the Last Meeting - Ctd.

Edge Length-Based Normality Tests

Edge Length-Based Normality Tests - Ctd.

Distribution of Edge Lengths - #1

edge-length-based-normality-test-20-percent.png
One distribution is only 20% the size of the other

Distribution of Edge Lengths - #2

edge-length-based-normality-test-40-percent.png
One distribution is only 40% the size of the other

Distribution of Edge Lengths - #3

edge-length-based-normality-test-60-percent.png
One distribution is only 60% the size of the other

Distribution of Edge Lengths - #4

edge-length-based-normality-test-80-percent.png
One distribution is only 80% the size of the other

Distribution of Edge Lengths - #5

edge-length-based-normality-test-identical-distributions.png
The distributions are identical

Distribution of Edge Lengths - #6

edge-length-based-normality-test-120-percent.png
One distribution is 120% the size of the other

Distribution of Edge Lengths - #7

edge-length-based-normality-test-140-percent.png
One distribution is 140% the size of the other

Distribution of Edge Lengths - #8

edge-length-based-normality-test-160-percent.png
One distribution is 160% the size of the other

Distribution of Edge Lengths - #9

edge-length-based-normality-test-180-percent.png
One distribution is 180% the size of the other

Distribution of Edge Lengths - #10

edge-length-based-normality-test-200-percent.png
One distribution is twice the size of the other

Another Look at the Histogram

Edge Length Distribution Simplified- #1

edge-length-based-histogram-20-percent.png
One distribution is only 20% the size of the other

Edge Length Distribution Simplified- #2

edge-length-based-histogram-identical.png
The distributions are identical

Edge Length Distribution Simplified- #3

edge-length-based-histogram-200-percent.png
One distribution is twice the size of another (that which remains unchanged in terms of size)

Euclidean Image Distance and the Effects of Noise

Specificity: Systematic Ratio Differences: Euclidean

Specificity-euclidean-with-colour.png

Generalisation: Systematic Ratio Differences: Euclidean

Generalisation-euclidean-with-colour.png

Specificity: Systematic Ratio Differences: Shuffle

Shuffle_5x5_specificity-warps-and-noise.png

Generalisation: Systematic Ratio Differences: Shuffle

Shuffle_5x5_generaliation-warps-and-noise.png

Discussion About the Results

Example Numbers (Results)


Experiment 6 (warp 370) Euclidean
Noise 0: Gen: 7.64902 se: 0.19764 Spec: 7.76829 se: 0.0304862
Noise 4: Gen: 7.91945 se: 0.199508 Spec: 7.96086 se: 0.0300599
Noise 8: Gen: 8.70313 se: 0.194002 Spec: 8.61653 se: 0.0286184
Noise 16: Gen: 11.7978 se: 0.195757 Spec: 11.5888 se: 0.0244965


Experiment 6 (wrap 370) Shuffle 5x5
Noise 0: Gen: 3.36806 se: 0.137414 Spec: 3.67181 se: 0.0196539
Noise 4: Gen: 3.59961 se: 0.150493 Spec: 3.7527 se: 0.0185894
Noise 8: Gen: 4.00571 se: 0.144773 Spec: 4.17225 se: 0.0171394
Noise 16: Gen: 5.71648 se: 0.115543 Spec: 6.11641 se: 0.0164059

Discussion About the Results - Ctd.

Sensitivity to Noise: A Glimpse at Results from Instantiation 6

Continued Discussion on Sensitivity

Noise Experiments and Inter-Instantiation Shuffle Distances

Generalisation - Inter-instantiation Variability - Shuffle

inter-instantiation-variability-test-Generalisation-to-noise-shuffle-window-5x5.png

Specificity - Inter-instantiation Variability - Shuffle

inter-instantiation-variability-test-Specificity-to-noise-shuffle-window-5x5.png

Generalisation - Inter-instantiation Variability - Euclidean

generalisablity-instantiation-5-and-6-overlaid.png

Specificity - Inter-instantiation Variability - Euclidean

specificity-instantiation-5-and-6-overlaid.png

Shuffle-based Localised Errors

Localised Errors in Non-Rigid Registration

37x37-average-shuffle-distance-image-window-5x5.png