brief: binary robust independent elementary features michael calonder, vincent lepetit, christoph...

19
BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Upload: noah-farrant

Post on 31-Mar-2015

231 views

Category:

Documents


5 download

TRANSCRIPT

Page 1: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

BRIEF: Binary Robust IndependentElementary Features

Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua

CVLab, EPFL, Lausanne, Switzerland

Page 2: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Contributions

• Pros:• Compact, easy-computed, highly discriminative• Fast matching using Hamming distance• Good recognition performance

• Cons:• More sensitive to image distortions and

transformations, in particular to in-plane rotation and scale change

Page 3: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Related work

• Descriptors: SIFT, SURF, DAISY, etc• Descriptor + Dimension Reduction (e.g.

PCA, LDA, etc)• Quantization• Hashing (e.g. Locality Sensitive Hashing)

Page 4: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Method

• Binary test

• BRIEF descriptor

• For each S*S patch1. Smooth it

2. Pick pixels using pre-defined binary tests

Page 5: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Smoothing kernels

• De-noising• Gaussian kernels

Page 6: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Spatial arrangement of the binary tests1. (X,Y)~i.i.d. Uniform

2. (X,Y)~i.i.d. Gaussian

3. X~i.i.d. Gaussian , Y~i.i.d. Gaussian

4. Randomly sampled from discrete locations of a coarse polar grid introducing a spatial quantization.

5. and takes all possible values on a coarse polar grid containing points

Page 7: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland
Page 8: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland
Page 9: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Distance Distributions

Page 10: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Experiments

Page 11: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

BRISK: Binary Robust Invariant Scalable Keypoints

Stefan Leutenegger, Margarita Chli and Roland Y. Siegwart

Autonomous Systems Lab, ETH Zurich

Page 12: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Contributions

• Combination of SIFT-like scale-space keypoint detection and BREIF-like descriptor• Scale and rotation invariant

Page 13: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Method

• Scale-space keypoint detection

Page 14: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

• Sampling pattern

Page 15: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

• Local gradient

• All sampling-point pairs

• Short-distance pairings S and long-distance pairings L

Page 16: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

• Overall characteristic pattern direction

• Descriptor• Rotation- and scale-normalization

• BRIEF-like

• Matching: Hamming distance

Page 17: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland

Experiments

Page 18: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland
Page 19: BRIEF: Binary Robust Independent Elementary Features Michael Calonder, Vincent Lepetit, Christoph Strecha, and Pascal Fua CVLab, EPFL, Lausanne, Switzerland