Computer Vision 연구자들이 알아야 할 20개 이상의 techniques

MS의 Richard Szeliski 박사님과 옥스포드의 Andrew Zisserman 교수님이 이야기하는 "모든 컴퓨터 비젼 연구자들이 알아야 할 20개 이상의 techniques"

 

1. Image formation and optics

2. Image processing, filtering, Fourier analysis

3. Pyramids and wavelets

4. Feature extraction

5. Image matching

6. Bag of words

7. Optical flow

8. Structure from motion

9. Multi view stereo

10.Segmentation

11.Clustering

12.Viola-Jones

13.Bayesian techniques

14.Machine learning

15.RANSAC and robust techniques

16.Numerical methods

17.Optimization

18.Range finding, active illumination

19.Algorithms

20.Graph cuts

21.Dynamic programming

22.Complexity analysis

23.MATLAB and C++. and assembly (optional: GPU programming)

24.Communication and presentation skills

 

Image and features

• NCC

• Interest point operators

• Scale invariant and affine invariant detectors & descriptors

• Scale space

• Image processing, filtering, Fourier analysis

• Pyramids and wavelets

• Edge detection

• Restoration e.g. deblurring, super-resolution

– Linear, e.g. Wiener filter

– MRF

– Non-local means/BM3D/bilateral filter

 

Segmentation, grouping and tracking

• Segmentation

– Normalized cuts

• Grouping

– Hough transforms

• Clustering

– K-means

– Mean-shift

– Pedro-clustering

• Tracking

– Kalmanfilter

– Particle filter

 

Multi-view: stereo, SFM, flow

• RANSAC and other robust techniques

• Geometry:

– epipolar geometry (projective and affine)

– planar homographies

– Affine camera

• Geometry estimators

– 8 point algorithm for F

– 4 point algorithm for H

• Factorization

• Bundle-adjustment

• Flow

– Horn & Schunck L2

– Lucas-Kanade

– L1 regularized

 

Recognition

• Bag of visual words

• HOG, SIFT, GIST

• Spatial pyramid

• Spatial configurations/Pictorial structures

• Sliding window/jumping window

• Cascades

​

Others

Machine Learning

– Adaboost

– kNN

– SVM

– Random forest

– PCA, ICA, CCA

– EM

– MIL/Latent-SVM

– Regularization

– HMM

– Graphical & Bayesian models

Optimization

– Classical linear and non-linear

– Graph operations

– Dynamic programming/message passing for MAP, max-marginals

– Graph cuts for binary variable MAP

• Texture synthesis