Tuesday, July 15, 2008

CS: a Video in Portuguese, a Summer School in French, a Tutorial in Switzerland, an MMDS talk and better 3d reconstruction.

Found on the internets:

A video entitled: Compressive Sampling: A new Paradigm in Signal Acquisition in Portuguese, actually it is Reconstrução de imagens subamostradas (compressed sensing) by Mário Figueiredo. It was added to the video section on Compressed Sensing.

A Summer school in France: École d'Été annuelle en traitement du signal et des images in the village of Peyresq will feature a presentation by Stéphane Chrétien entitled: Un decodeur 'two stage' pour le compressed sensing ( A two-stage decoder for Compressed Sensing).

Résumé : Le probleme de compressed sensing de Candes et Tao consiste a retrouver un signal qui admet une representation parcimonieuse dans un dictionnaire connu au moyen d'un nombre limite de mesures. Le decodeur de plus employe actuellement est la relaxation l1 du probleme combinatoire NP-difficile consistant a chercher le vecteur le plus sparse satisfaisant un systeme de contraintes affines. Nous proposons d'etudier une nouvelle relaxation Two Stage tres efficace pour le probleme de compressed sensing. Cette relaxation est aussi performante en pratique que l'approche dite Reweighted l1 de Candes, Wakin et Boyd (Journal of Fourier Analysis and Applications, a paraitre) mais a l'avantage d'etre gouvernee par un parametre de reglage a l'interpretation "physique" claire et tres facile a choisir. Si le temps le permet, on montrera comment obtenir des relaxations plus fines encore par dualite lagrangienne.

It looks like an extension of his paper: An Alternating l_1 Relaxation for compressed sensing.


At EUSIPCO, there will be a tutorial session. On Monday, August 25th from 2:30 pm to 5:50 pm, one of them will feature: Theory and Applications of Compressive Sensing by Richard Baraniuk. The smmary reads:

Sensors, cameras, and imaging systems are under increasing pressure to accommodate ever larger and higher-dimensional data sets; ever faster capture, sampling, and processing rates; ever lower power consumption; communication over ever more difficult channels; and radically new sensing modalities. The foundation of today's digital data acquisition systems is the Shannon/Nyquist sampling theorem, which asserts that to avoid losing information when digitizing a signal or image, one must sample at least two times faster than the signal's bandwidth, at the so-called Nyquist rate. Unfortunately, the physical limitations of current sensing systems combined with inherently high Nyquist rates impose a performance brick wall to a large class of important and emerging applications. In digital image and video cameras, for instance, the Nyquist rate is so high that too many samples result, making compression by algorithm like JPEG or MPEG a necessity prior to storage or transmission. In imaging systems (medical scanners and radars) and high-speed analog-to-digital converters, increasing the sampling rate is very expensive or detrimental to a patient's health.

Compressive Sensing is a new approach to data acquisition in which analog signals are digitized for processing not via uniform sampling but via measurements using more general, even random, test functions. In stark contrast with conventional wisdom, the new theory asserts that one can combine "low-rate sampling" with digital computational power for efficient and accurate signal acquisition. Compressive sensing systems directly translate analog data into a compressed digital form; all we need to do is "decompress" the measured data through an optimization on a digital computer. The implications of compressive sensing are promising for many applications and enable the design of new kinds of analog-to-digital converters, cameras, and imaging systems.

This tutorial will overview the theory of compresive sensing, point out the important role played by the geometry of high-dimensional vector spaces, and discuss how the ideas can be applied in next-generation acquisition devices. Particular topics include sparse signal representations, convex optimization, random projections, the restricted isometry principle, the Johnson-Lindenstrauss lemma, Whitney's embedding theorem for manifolds, and applications to imaging systems, sensor networks, and analog-to-digital converters.

Here is a new technique competing with Make3D previously mentioned here before. It is featured in Closing the Loop on Scene Interpretation by Derek Hoiem, Alexei A. Efros, Martial Hebert. Here is a 3D reconstruction video comparing it with the older Photo Pop-up and Make3D. Unlike Make3D, the software is not available, so we can't really check the issue of out-of-this world renderings.


A new presentation at MMDS 2008. Workshop on Algorithms for Modern Massive Data Sets just showed up. It is that of Inderjit Dhillon entitled Rank Minimization via Online Learning.

Credit: NASA/JPL-Caltech/University of Arizona/Texas A&M. Poppy fields, sol 49.

2 comments:

Anonymous said...

Hi,

I have a question about:

"Rank Minimization via Online Learning" by Inderjit Dhillon

It was very interesting presentation, is it based on a paper/report submitted to the workshop? if yes, How can I have access to that paper?

It would be great you can cover this workshop too:
http://www.msri.org/calendar/sgw/WorkshopInfo/451/show_sgw

I think it is very related.

Regards,
Kayhan

Igor said...

Hello Kayhan

It looks like the paper is here:
http://www.cs.utexas.edu/~inderjit/public_papers/rankmin_icml.pdf

reachable from Inderjit Dhillon page.

Cheers,

Igor.

Printfriendly