linear solution to scale and rotation invariant object matching professor: 王聖智 教授 student...

39
Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王王王 王王 Student : 王

Post on 15-Jan-2016

232 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Linear Solution to Scale and Rotation Invariant Object Matching

Professor: 王聖智 教授 Student : 周 節

Page 2: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Outline Introduction Method Result Conclusion Reference

Page 3: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Problem

Template image Target image

3

Page 4: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Problem

Template image and mesh

Target image

Matching

Target Mesh

Page 5: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Challenges

Template Image

Target Images

Page 6: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Challenges

Template Image

Target Images

Page 7: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

:wRelated Works

Hough Transform and RANSAC Graph matching

Dynamic Programming Max flow – min cut [Ishikawa 2000, Roy 98] Greedy schemes (ICM [Besag 86], Relaxation

Labeling [Rosenfeld 76]) Back tracking with heuristics [Grimson 1988] Graph Cuts [Boykov & Zabih 2001] Belief Propagation [Pearl 88, Weiss 2001]

Convex approximation [Berg 2005, Jiang 2007, etc.]

Page 8: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Method

1. A linear method to optimize the matching from template to target

2. The linear approximation is simplified so that its size is largely decoupled from the number of target candidate features points

3. Successive refinement for accurate matching

Page 9: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Method

1. A linear method to optimize the matching from template to target

2. The linear approximation is simplified so that its size is largely decoupled from the number of target candidate features points

3. Successive refinement for accurate matching

Page 10: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

The Optimization Problem

Page 11: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

The Optimization Problem

Matching cost Pairwise consistency

Page 12: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

In Compact Matrix Form

Matching cost Pairwise consistency

Page 13: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

In Compact Matrix Form

We will turn the terms in circles to linear approximations

Matching cost Pairwise consistency

Page 14: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

The L1 Norm Linearization

It is well known that L1 norm is “linear”

min |x| min (y + z)Subject to x = y - z y, z >= 0

Page 15: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

The L1 Norm Linearization

It is well known that L1 norm is “linear”

By using two auxiliary matrices Y and Z,

| EMR – sEXT | 1’ne ( Y + Z ) 12

subject to Y – Z = EMR – sEXT Y, Z >= 0

Consistency term

Consistency term

min |x| min (y + z)Subject to x = y - z y, z >= 0

Page 16: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Linearize the Scale Term

For any given sCut into Ns part 0 <S1<. . .< Sns

Page 17: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Linearize the Scale Term

For any given sCut into Ns part 0 <S1<. . .< Sns

For any given XX1, X2…Xns

Page 18: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Linearize the Scale Term

For any given sCut into Ns part 0 <S1<. . .< Sns

For any given XX1, X2…Xns

Page 19: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Linearize the Scale Term

For any given sCut into Ns part 0 <S1<. . .< Sns

For any given XX1, X2…Xns

Page 20: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Linearize the Scale Term

All sites select the same scale

Page 21: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Linearize Rotation Matrix

In graphics:

Page 22: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

The Linear Optimization

T* = XT

Target point estimation:

Page 23: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Method

1. A linear method to optimize the matching from template to target

2. The linear approximation is simplified so that its size is largely decoupled from the number of target candidate features points

3. Successive refinement for accurate matching

Page 24: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Convex optimization problem

24

Reference:http://www.stanford.edu/class/ee364a/lectures/problems.pdf

Page 25: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

The Lower Convex Hull Property Cost surfaces can be replaced by their lower convex

hulls without changing the LP solution

A cost “surface” for one point on the template

The lower convex hull

Page 26: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Removing Unnecessary Variables Only the variables that correspond to the lower convex

hull vertices need to be involved in the optimization

# of original target candidates: 2500

# of effective target candidates: 19

Page 27: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Complexity of the LP

28 effective target candidates 1034 effective target candidates

10,000 target candidates 1 million target candidates

The size of the LP is largely decoupled from the number of target candidates

Page 28: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Method

1. A linear method to optimize the matching from template to target

2. The linear approximation is simplified so that its size is largely decoupled from the number of target candidate features points

3. Successive refinement for accurate matching

Page 29: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Successive Refinement

Page 30: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Result

Page 31: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Matching Objects in Real Images

Page 32: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Result Videos

Page 33: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Statistics

book magazine bear butterfly bee fish

#frames 856 601 601 771 101 131

#model 151 409 235 124 206 130

#target 2143 1724 1683 1405 1029 7316

time 1.6s 11s 2.2s 1s 2s 0.9s

accuracy 99% 97% 88% 95% 79% 95%

Page 34: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Results Comparison

Page 35: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Accuracy and Efficiency

Page 36: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Experiments on Ground Truth Data

Error distributions for fish dataset

The proposed method

Othermethods

Page 37: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Experiments on Ground Truth Data

Error distributions for random dot dataset

The proposed method

Othermethods

Page 38: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Conclusion

A linear method to solve scale and rotation invariant matching problems accurately and efficiently The proposed method is flexible and can be used to

match images using different features It is useful for many applications including object

detection, tracking and activity recognition

Page 39: Linear Solution to Scale and Rotation Invariant Object Matching Professor: 王聖智 教授 Student : 周 節

Reference

Hao Jiang and Stella X. Yu, “Linear Solution to Scale and Rotation Invariant Object Matching”, IEEE Conference on Computer Vision and Pattern Recognition 2009 (CVPR'09)

http://www.stanford.edu/class/ee364a/lectures.html