Old course 048926, 2013-2014 

Course given in fall semester 2013-2014.

Course syllabus and info (Hebrew & English)

Lectures

Week

Topic

Slides

Notes

1 Intro lecture Intro  
2 Diffusion processes I- linear, Perona-Malik,Weickert Week 2 Notes 1
3 Diffusion processes II - some analysis andnumerics   Notes 2
4 Functionals I Week 4 Notes 3
5 BV and Euler-Lagrange Week 5 Notes 4
6 TV Denoising anddeconvolution Week 6 Notes 5
7 Convexity and TVnumerics   Notes 6
8 Segmentation I Week 8 Notes 7
9 Segmentation and Level-Sets Week 9 Notes 8
10 Optical flow Week 10 Notes 9
11 Image decomposition   Notes 10
12 Nonlocal operators Week 12 Notes 11
13 Numerical methods   Notes 12

 

Exercises

Ex#

Topic

PDF

Due date

Files

1 Diffusion Ex1 18.11.2013 Ex1_files
2 TV Denoising and Deconvolution Ex2 16.12.2013  
3 Segmentation and optical flow Ex3 13.1.2014 Images
4 Nonlocal operators Ex4 24.2.2014 Images
Proj Final project Final Proj 24.2.2014 See paper list below

Project papers

Paper#

Topic

Authors

PDF

1 Segmentation – active contours Chan, Vese Paper1
2 Motion and stereo Valgaerts, Bruhn et al Paper2
3 Segmentation – nonlocal Jung, Peyre, Cohen Paper3
4 Optical flow Xu, Jia, Matsushita Paper4
5 Segmentation  - Texture Rousson, Brox, Deriche Paper5
6 Deblurring – impulse noise Bar, Brook, Sochen,Kiryati Paper6
7 Segmentation – fast global optimization Bresson, Esedoglu  et al Paper7
8 Depth cameras Kerl, Sturm, Cremers Paper8
9 Segmentation – shape priors Cremers, Osher, Soatto Paper9
10 Stereo with occlusions Ben-Ari, Sochen Paper10
11 Optimal denoising Gilboa, Sochen, Zeevi Paper11
12 Nonlocal operators Gilboa, Osher Paper12

 

 

NEW

Lecture on 20.1.2014 canceled

The lecture on 20.1.2014 is canceled, instead there will be a project presentation meeting or 17.2.2014 (same hour, same room).

Note that on 20.1 there is Israeli Computer Vision Day the yearly Israeli meeting on image processing and computer vision in Herzeliya (free entrance).

 

For the interested readers, papers related to Ex 3

·        Matlab-Q1. Ambrosio and Tortorelli

·        Matlab-Q2. Brox et al

 

Older messages:

Comments about Ex-3

·        In Analytic 1.(b) and 1.(c) assume a very simple solution. Some solutions are degenerate (and do not produce a segmentation) with alpha=1, beta=1. With smaller beta one receives more “logical” solutions.

·        Matlab Q1. Ambrosio-Tortorelli – one should derive the E-L equations, one for u, one for v. Note that when deriving for u, v is a function v(x), this should be taken into account when doing the derivative with respect to u’.

·        Matlab Q2. One needs to compute it using a coarse to fine implementation. Basically building a Guassian pyramid of the images (smaller scales by factors of 2,4,8..). Find the optical flow in the coarse scale (where the maximal shift is around 1 pixel) and propagate this flow down the pyramid to the next finer scale (either by using the coarse flow as initial condition for the minimization or by moving (warping) the second image according the coarse estimated flow).

Ex-2 delayed to 16.12

Due to several requests and in order for you to have more free time during Hanukah, the deadline for Exercise 2 is delayed by a week to 16.12.2013.

 

Files for Exercise 1 were uploaded

Changes in Ex-1 (analytic part):

1.    Q1 (b) – only translation and rotation invariance.

2.    Q2 (b) (i) – a more precise explanation is given.

PDF of Ex1 is updated with these changes.