40
Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Embed Size (px)

Citation preview

Page 1: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Shape Matching

and Object Recognition

using Low Distortion Correspondence

Alexander C. Berg, Tamara L. Berg, Jitendra Malik

U.C. Berkeley

Page 2: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Object Category Recognition

Page 3: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Deformable Template Matching with Exemplars for Recognition

● Use exemplars as deformable templates

● Find a correspondence between the query image and each template

Query

Image

Database of

Templates

Page 4: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Deformable Template Matching with Exemplars for Recognition

● Use exemplars as deformable templates

● Find a correspondence between the query image and each template

Query

Image

Database of

Templates

Best matching template is a helicopter

Page 5: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence for Deformable Template Matching

● Evaluate correspondence based on:

– Similarity of appearance near feature points

– Similarity in configuration of the feature points

QueryTemplate

Page 6: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence for Deformable Template Matching

● Evaluate correspondence based on:

– Similarity of appearance near feature points

– Similarity in configuration of the feature points

QueryTemplate

Page 7: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence for Deformable Template Matching

● Evaluate correspondence based on:

– Similarity of appearance near feature points

– Similarity in configuration of the feature points

QueryTemplate

Page 8: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence for Deformable Template Matching

● Evaluate correspondence based on:

– Similarity of appearance near feature points

– Similarity in configuration of the feature points

QueryTemplate

Page 9: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Result

QueryTemplate

Page 10: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Interpolated CorrespondenceUsing Thin Plate Splines

QueryTemplate

Page 11: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence for Deformable Template Matching

If i i' and j j' then rijri'j'

If i i' then pipi'

i i'

j j'

i i'

j j'QueryTemplate r

ijr

i'j'

Page 12: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Geometric Blur(Local Appearance Descriptor)

Geometric Blur Descriptor

~

Compute sparse

channels from image

Extract a patch

in each channel

Apply spatially varying

blur and sub-sample

(Idealized signal)

Descriptor is robust to small affine distortions

Berg & Malik '01

Page 13: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Geometric Blur(Local Appearance Similarity)

Geometric Blur Descriptor

Geometric Blur Descriptor~

Page 14: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Are Features Enough?

Not Quite...

Color indicates

similarity using

Geometric Blur

Descriptor

Page 15: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Measuring Distortion(Similarity in Configuration)

i i'

j j'

i i'

j j'QueryTemplate R

ij 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

Page 16: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Cost Function as IQP

Appearance costif i -> j

Distortion cost ifi -> j and k -> l

Integer Quadratic Programming Problem...

iff template point i maps to query point j, If binary vector x represents a correspondence

cf. Maciel & Costeira '03

Page 17: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Optimization

● 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.

Page 18: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Result

Page 19: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Interpolated CorrespondenceUsing Thin Plate Splines

Page 20: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Quadratic Assignment(Using IQP)

Page 21: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Linear Assignment(e.g. Hungarian)

Page 22: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples (Shape Matching)

Page 23: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 24: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 25: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 26: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 27: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 28: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 29: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 30: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 31: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 32: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Correspondence Examples(Shape Matching)

Page 33: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Application to RecognitionCaltech 101

– 101 classes of objects + “background”

● Large Scale

● Roughly aligned

● Large intra-class variation

● Fei-Fei, Fergus, Perona '04

Page 34: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Caltech 101 Recognition Results

Chance ~1%

N.N. whole image 16%

Discriminative version ofConstellation Model 27%

N.N. Geometric BlurDescriptors 38%

Low Distortion Correspondence (GB+IQP) 45%

102 way Alternative Forced Choice test

(15 training examples per class)

102 way confusion matrix

100%

0%

Page 35: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Model Building for Segmentation

Rough correspondenceto each example image

Average quality of alignment

Threshhold

Page 36: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Automatic vs Hand Segmentation

Page 37: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Application 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

Page 38: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Application to RecognitionFaces

Page 39: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Conclusion

● 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

Page 40: Shape Matching and Object Recognition using Low Distortion Correspondence Alexander C. Berg, Tamara L. Berg, Jitendra Malik U.C. Berkeley

Thank You

Acknowledgements

Charless FowlkesXiaofeng RenDavid Forsyth