1. Study and application of superresolution techniques on the OSIRIS cameras onboard Rosetta
Khallefi Leïla
© esa
Supervisors:
J. L. Vazquez
M. Küppers
trainee project 2009

2. Mission Instrument Superresolution Goals My Work
Agenda - Mission - Instrument - SR - Goals - My Work
Mission
Instrument
Superresolution
Goals
My Work

3. A key in the comprehension of the solar system formation
Agenda - Mission - Instrument - SR - Goals - My Work
Rosetta Mission
Why studying comets?
A key in the comprehension of the solar system formation
3rd Cornerstone Mission of the ESA Horizon 2000 Program
© esa
© esa
Launch 2004 – encounter 2014
Target Comet Churyumov-Gerasimenko
Rosetta Space Probe
Lander Philae
Asteroid Steins

4. OSIRIS Optical, Spectroscopic and Infrared Remote Imaging System
Agenda - Mission - Instrument - SR - Goals - My Work
OSIRIS
Optical, Spectroscopic and Infrared Remote Imaging System
WAC wide angle camera
NAC narrow angle camera
Provide images of the near nucleus environnement
14 filters between nm
Resolution of 101 µrad px-1
Provide High resolution images
12 filters
Resolution of 18.6 µrad px-1
Shape and volume
Surface mineralogy
Landing spot for Philae
Nature of the cometary nuclei …
Provide information on :

5. How to find a way to increase the current resolution ?
Superresolution
Agenda - Mission - Instrument - SR - Goals - My Work
How to find a way to increase the current resolution ?
Limitations
Maintain the complete nucleus within the FOV of the WAC
Mass requirements for a longer Focal length system
Limited the data volume that can be transmitted
Cost of the CCD
© xkcd
A solution: Image processing techniques

6. Exploit the new information in each image
Superresolution
Agenda - Mission - Instrument - SR - Goals - My Work
Exploit the new information in each image
Multiple images
scene …
spacecraft
Subpixel Shifts
Aliasing
Motion estimation
S. Park, M. Park, M. Kang
IEEE signal processing magazine may 2003
spacecraft
scene …

7. Study and comparison of algorithms
Agenda - Mission - Instrument - SR - Goals - My Work
I) Bibliography
Study and comparison of algorithms
II) Implementation of algorithms /tests
III) Use on the images of the Asteroid Steins
IV) Specifications on images
V) Search engine development

8. Step one Study and Comparision construction of a model:
Agenda - Mission - Instrument - SR - Goals - My Work
Step one Study and Comparision
construction of a model:
Wk X + nk =
Wk X X ?
sampling
Continuous to Discrete Without Aliasing
Continuous scene
HR image x
Kth Observed LR image
warping blurring
downsampling
Translation Rotation
Blur
noise +

9. Step one Study and Comparision construction of a model:
Agenda - Mission - Instrument - SR - Goals - My Work
Step one Study and Comparision
construction of a model:
the HR image
Yk = Wk X + nk
noise
the kth LR image
Warping blurring subsampling
Matrix gathers
- estimation of the motion
- noise statistics
- Sensing
S. Park, M. Park, M. Kang, IEEE signal processing magazine may 2003
knowledge

10. Step one Study and Comparision
Agenda - Mission - Instrument - SR - Goals - My Work
Step one Study and Comparision
6 main different approaches
Non uniform Interpolation
More sophisticated models for registration error (than just gaussian)
Differents weights according to the contribution of the LR images
2 algorithms
+ improvement:
Frequency domain
+ Not constrain noise and blur
+ Reduce effect of registration error
+ Theoretically simple
Lack of correlation in the frequency domain
Regularized reconstruction
+ Flexible
+ Use of prior knowledge
+ Use when small number of LR images available
Frequency domain
+ Not constrain noise and blur
+ Reduce effect of registration error
+ Theoretically simple
Lack of correlation in the frequency domain
Projection onto convex sets
+ Simple
- Non uniqueness of the solutions
- High computational cost
Adaptative filtering
+ Really efficient
- Computational complexity
Regularized reconstruction
+ Flexible
+ Use of prior knowledge
+ Use when small number of LR images available
Iterative back projection
+ simplicity
- Non uniqueness of the solution
- Choice of the parameter
+ Real time
- Same noise and blur in all the images
- Ignore error of interpolation stage

11. Step two Implementation
Agenda - Mission - Instrument - SR - Goals - My Work
Step two Implementation
The Bayesian Approach
compute the MAP grid
use a prior (smooth)
estimation of parameters of regularisation
use an image for reference
interpolate pixels values at the desired resolution
test
HR image

12. Step two Implementation
Agenda - Mission - Instrument - SR - Goals - My Work
Step two Implementation
The Bayesian Approach
S. Park, M. Park, M. Kang, IEEE signal processing magazine may 2003
step 1: use an image
- images of same filter?
- resize them
- estimation of the motion with spice
- choosing the resolution?
- create a new picture by interpolation
- interpolation method coherent with the sensing: bilineare, Spline,…
- opening/closing - using of spice center of mass
- Type of motion estimation or knowledge

13. Step two Implementation
Agenda - Mission - Instrument - SR - Goals - My Work
Step two Implementation
The Bayesian Approach
step 2: iterative process
- estimation of the parameters
By minimisation of the square error
- creation of a new image
- re-evaluation of the parameters
Issues with band choices
Number of images used
Issue with the bond
choice of the interpolation process
Size of the interpolation grid
Number of iterations
Estimation of the noise, modification

14. Agenda - Mission - Instrument - SR - Goals - My Work
TO BE CONTINUED…