Group MembersIbrahim Sadek

Mohamed Elawady

1

Robust principal axes determination for point-based

shapes

Overview

Framework

Discussion & Results

Conclusion & Future Work

Agenda

2

Overview

Use (Least Median of Squares)

instead of (Least Mean of Squares)3

Major & Minor Regions

Framework

Calculate principal component axes based on LMS & forward search technique

Extract the initial subset for forward search technique using octree-based approximation

Apply the proposed algorithm to some shape matching techniques with/without the effect of noise

4

Framework

LMS RPA

5

Discussion & Results

6

Discussion & Results

7

Run #1 Run #2 Run #3

Comparison of major regionsCommon (Chest, Abdomen)

Discussion & Results

8

Dog Human Horse

Comparison of major regionsCylinder-like Shape

Discussion & Results

9

Comparison of major regionsWith/Without Noise

Run #1Normal

Run #2Cropping

Run #3Gaussian Noise

Discussion & Results

10

Comparison of major regionsWith/Without Noise

Run #1 Run #2 Run #3

Discussion & Results

3D Mesh Operations

PCA Computation

Octree Operations

3D Shape Matching

11

Problems

Conclusion

12

Shape Noise PCABad

(Shape Matching)

Shape Noise RPCAGood

(Shape Matching)

All points PAs

Major points PAs

Future Work

13

• IN PROGRESS (Not Sure About Results!!)Fixing E2 & E3

• Needs Training !!

• Depends on dataset shapes

Choosing Correct

Parameters

• Iterative Cloud Points

• Transformation Matrix (depends on O & E1)

Shape Matching

14

Questions?!15