Title: Shape Matching and Object Recognition using Low Distortion Correspondence
1Shape Matching and Object Recognition using Low
Distortion Correspondence
- Alexander C. Berg, Tamara L. Berg, Jitendra
Malik - U.C. Berkeley
2Object Category Recognition
3Deformable Template Matching with Exemplars for
Recognition
- Use exemplars as deformable templates
- Find a correspondence between the query image and
each template
Database of Templates
Query Image
4Deformable Template Matching with Exemplars for
Recognition
- Use exemplars as deformable templates
- Find a correspondence between the query image and
each template
Database of Templates
Query Image
Best matching template is a helicopter
5Correspondence for Deformable Template Matching
Query
Template
- Evaluate correspondence based on
- Similarity of appearance near feature points
- Similarity in configuration of the feature points
6Correspondence for Deformable Template Matching
Query
Template
- Evaluate correspondence based on
- Similarity of appearance near feature points
- Similarity in configuration of the feature points
7Correspondence for Deformable Template Matching
Query
Template
- Evaluate correspondence based on
- Similarity of appearance near feature points
- Similarity in configuration of the feature points
8Correspondence for Deformable Template Matching
Query
Template
- Evaluate correspondence based on
- Similarity of appearance near feature points
- Similarity in configuration of the feature points
9Correspondence Result
Query
Template
10Interpolated CorrespondenceUsing Thin Plate
Splines
Query
Template
11Correspondence for Deformable Template Matching
Query
Template
ri'j'
rij
12Geometric Blur(Local Appearance Descriptor)
Berg Malik '01
Compute sparse channels from image
Extract a patch in each channel
Apply spatially varying blur and sub-sample
Descriptor is robust to small affine distortions
Geometric Blur Descriptor
(Idealized signal)
13Geometric Blur(Local Appearance Similarity)
Geometric Blur Descriptor
Geometric Blur Descriptor
14Are Features Enough?
Color indicates similarity using Geometric Blur
Descriptor
Not Quite...
15Measuring Distortion(Similarity in Configuration)
Query
Template
Rij
Si'j'
Measure distortion in vectors between pairs of
feature points - R and S same length for
rotations - R and S same direction for scalings
16Cost Function as IQP
cf. Maciel Costeira '03
Appearance cost if i -gt j
Distortion cost if i -gt j and k -gt l
Integer Quadratic Programming Problem...
17Optimization
- Integer Quadratic Programming is NP hard
- The instances we generate seem easy
- Using a linear bound to initialize gradient
descent provides good results - (In fact better than the guarantee of Goemans
Williamson's randomized algorithm) - Varying the linear constraints on x allows
- one-one, one-many, or fixed number of outliers,
etc.
18Correspondence Result
19Interpolated CorrespondenceUsing Thin Plate
Splines
20Quadratic Assignment(Using IQP)
21Linear Assignment(e.g. Hungarian)
22Correspondence Examples (Shape Matching)
23Correspondence Examples(Shape Matching)
24Correspondence Examples(Shape Matching)
25Correspondence Examples(Shape Matching)
26Correspondence Examples(Shape Matching)
27Correspondence Examples(Shape Matching)
28Correspondence Examples(Shape Matching)
29Correspondence Examples(Shape Matching)
30Correspondence Examples(Shape Matching)
31Correspondence Examples(Shape Matching)
32Correspondence Examples(Shape Matching)
33Application to RecognitionCaltech 101
- 101 classes of objects background
- Large Scale
- Roughly aligned
- Large intra-class variation
- Fei-Fei, Fergus, Perona '04
34Caltech 101 Recognition Results
102 way Alternative Forced Choice test (15
training examples per class)
Chance 1 N.N. whole image 16 Discrimi
native version of Constellation Model 27 N.N.
Geometric Blur Descriptors 38 Low
Distortion Correspondence (GBIQP) 45
102 way confusion matrix
100
0
35Model Building for Segmentation
Average quality of alignment
Rough correspondence to each example image
Threshhold
36Automatic vs Hand Segmentation
37Application to RecognitionFaces
- Face dataset from Berg et al '03
- Medium to large scale faces
- AP News photos
- 20 face exemplars
- Same methodology as Caltech 101, but multiple
objects / image - After one face is identified its features are
removed and the search continues - Compared to a detector from Mikolajczyk based on
Schneiderman Kanade, that is quite successful
on this dataset
38Application to RecognitionFaces
39Conclusion
- Use rich descriptors that are insensitive to
typical transformations - Geometric Blur
- Enforce relationship constraints among
corresponding features - Integer Quadratic Programming
- Estimate smooth transform
- Thin Plate Splines
40Thank You
Acknowledgements Charless Fowlkes Xiaofeng
Ren David Forsyth