LSB Galaxies Detection Using Markovian Segmentation on Astronomical - - PowerPoint PPT Presentation

LSB Galaxies Detection Using Markovian Segmentation on Astronomical Images

Mireille Louys1, Benjamin Perret 1 Bernd Vollmer 2, François Bonnarel 2 Sebastien Lefèvre1, Christophe Collet1

1 Laboratoire des Sciences de l’Informatique , de l’Image et de la Télédétection 2

Centre de donnees astronomiques de Strasbourg,

Rationale

• Surface brightness is highly variable from one

galaxy to another (Disney, 1976)

• Low surface brightness galaxies are close to

the image background level and satisfy: SB > 22.5 mag.arcsec-2

• Found inside and outside galaxy clusters

– How complete are our detections?

• More understanding in the baryonic fraction of

the Dark Matter in the universe

2

Noise and background

3

Easy case More difficult

Overview

Markovian Quadtree

Segmentation Objects Profiles Fitting Selection DetectLSB

Ellipse parameters Fit error Magnitude profile LinLog, LinLin Measured Surface brightness Estimated Surface brightness Scale length

Visual Inspection and Validation

3D mathematical morphology Segmentation Map Original Image Objects list VO Objects list

VOTable XML

Markovian segmentation

• The goal is to classify pixel values and discriminate

bright objects, sky background and faint sources to be considered later as LSB candidates .

• It works as an inverse method :

– Try to find the Bayesian estimate of the most likely class partition that maps the observation – We make 2 assumptions :

• on the noise and inner class statistics (here gaussian),
• on the statistical phenomenon: Markovian quadtree.

5

Markovian quadtree

• The succesive tree levels correspond

to increasing scales of the image.

• We want to compute the a posteriori

probability of having site s in one class c, knowing the observation Ys and the class number of the parent node within the Markov tree.

• The Quadtree provides an in-scale

regularisation framework, providing exact and non iterative solution

• It is fully applicable to multiband
• bservations

6

Steps achieved with DetectLSB

• For each detection map:

– Identify pixel connected components – Remove very extended components : > 300 pixels – Remove components at the edges of the image

• Ellipse fitting
• Luminosity profile analysis
• Selection criteria:

– Sort out stars according to their profile (steep slope, etc.) – Remove bad centered and/or overlaping objects (crowded regions) – Check for central surface brightness

7

Luminosity profile analysis

• For each object in the detection map:

– Average profile along each ellipse radius

f(r)=U0*exp(‐r/R0) At least 3 points aligned above background level

Source Measurements

Surface Brightness Profile Fit Error Radius Fitted SB

Data set

• B-images of a region of the Virgo cluster
• btained with the Isaac Newton Telescope

WFCS (Wild Field Camera Survey)

• Analysed by Sabina Sabatini et coll. in

SABATINI S.; DAVIES J.; SCARAMELLA R.; SMITH R.; BAES M.; LINDER S.M.; ROBERTS S.; TESTA V. Mon. Not.

• R. Astron. Soc., 341, 981-992 (2003) .

and provided to us as a test set

• Full data set = 80 images of 4096x2048 pixels
• Analysed: 18 images with X-match with the

detection lists provided by the authors

10

Results on the INT data set

• Found 79 % of our LSB candidates are

confirmed by Sabatini et al.

23 (Markov+detectLSB) compared to 29 (Sabatini)

• Found new LSB candidates
• The validation on the full data set is still an
• n-going collaboration effort with S. Sabatini

and W. van Driel.

11

Conclusion

• The Markovian segmentation approach allows to

study LSB candidates even in a noisy environment

• Multiband image analysis is beeing currently applied

to the same data set in B band completed with I band

• First tests are promising
• Such a procedure has been implemented in the AIDA

image processing workflow project at CDS (Cf Schaaff et al, this conference)

12

2 detections for the same

• bject