Presentation is loading. Please wait.

Presentation is loading. Please wait.

The Role of Computer Vision in Astronomy

Similar presentations


Presentation on theme: "The Role of Computer Vision in Astronomy"— Presentation transcript:

1 The Role of Computer Vision in Astronomy
Rob Fergus New York University Pietro asked to speak briefly about how computer vision is being used in astronomy.

2 Overview Virtually all our knowledge about the universe derived from measurements of photons Usually as images Big astronomy project is $50M-200M But only 1-2% of this on software Just discovering techniques from computer vision & machine learning Now these are two areas which we don’t think of as being closely related. But its worth realizing that the vast majority of our knowledge of the universe comes from looking at images of the sky. Correspondingly, astronomers spend serious money to capture the best images – space-based telescopes >$1Bn. But the interesting thing from our point of view is that they just don’t have the same emphasis on software. However, this is changing as we can discuss later. As a community, they are just discovering methods from vision and ML and so there is a lot of low-hanging fruit from our perspective.

3 Astrometry.net Input: image of sky Output:
Absolute position List of objects Geometric hashing (quads of stars) Lamdan & Wolfson [ICCV’88] Widely used by pros & amateurs Lang, Hogg, Mierle, Blanton & Roweis, [The Astronomical Journal, Vol. 137, 2010]

4 Removing Atmospheric Distortions
Ground-based telescopes look through atmosphere Blind (online) estimation of atmospheric distortion and true image Far better than “lucky” imaging (current approach) Hirsch, Harmeling, Sra & Schölkopf , [Astronomy & Astrophysics 2011] Hirsch, Sra, Schölkopf & Harmeling, [CVPR 2010]

5 Exoplanet Imaging Want to image planets around other stars
Need contrast ratio >1010 for Earth-like planets Diffraction in telescope Light from star obscures planet Deconvolution problem Big assistance from optical design Planet Zimmerman et al., [Astrophysical Journal, 2009]. Big assistance from  Plus clever Oppenheimer and Hinkley, [Annual Review of Astronomy and Astrophysics, 2009]. Crepp et al., [Astrophysical Journal, Vol. 729, 2011].

6 Galaxy / Star Classification
Star vs Galaxy [Sloan Digital Sky Survey]

7 Galaxy / Star Classification
Distinguish stars from galaxies SVM-based models Smith et al. [A & A, Vol. 522, 2010] Generative model of galaxies Lang et al. [In preparation] Stars Galaxies Data Model

8 Future Directions Unified Bayesian model
Propagate uncertainty from pixels Physics- informed priors Low signal to noise regime is typical Already use Bayesian modeling for high-level cosmological models Data is far simpler than natural images Funded by NSF CDI

9

10 Cosmology Bayesian approaches to fitting high-level cosmological models

11 Cosmic Ray Classification
Raw image from Hubble Space Telescope:


Download ppt "The Role of Computer Vision in Astronomy"

Similar presentations


Ads by Google