use photutils segmentation image to set size of cutout

[rfinn]

enough said

[rfinn]

so where best to implement this? We need to run photutils

https://photutils.readthedocs.io/en/stable/segmentation.html

from photutils import detect_threshold
threshold = detect_threshold(data, nsigma=2.)

detect sources. do we want to use a kernel?

from astropy.convolution import Gaussian2DKernel
from astropy.stats import gaussian_fwhm_to_sigma
from photutils import detect_sources
sigma = 3.0 * gaussian_fwhm_to_sigma  # FWHM = 3.
kernel = Gaussian2DKernel(sigma, x_size=3, y_size=3)
kernel.normalize()
segm = detect_sources(data, threshold, npixels=5, filter_kernel=kernel) 

Incorporate deblending?

from photutils import deblend_sources
segm_deblend = deblend_sources(data, segm, npixels=5,
...                                filter_kernel=kernel, nlevels=32,
...                                contrast=0.001)

Then find galaxy within segmentation image

Then fit ellipse to shape, or just use x and y min/max of segmentation image.

from photutils import source_properties
cat = source_properties(data, segm_deblend)

we can get the bounding box from the source_properties

bbox_xmax bbox_xmin bbox_ymax bbox_ymin centroid - returns (y,x) of the data

[rfinn]

We could then expand the bounding box by a factor of e.g. 1.5.

[rfinn]

We could make the segmentation image from the r-band image, and then save the catalog. When the galaxies are selected, we could get the segmentation data values at the position of the galaxy. this should give the object number of the galaxy in the segmentation image. Then grab values from table with corresponding row.

[rfinn]

I tried both photutils and source extractor segmentation images. source extractor is much faster - like 2 seconds compared to 30 for photutils. however, source extractor included some far flung pixels in the segmentation image of one galaxy. This ended up increasing the cutout size, and since I don't know what is causing it, it's hard to know how many images will be affected and by how much.

so in the end I decided to stick with photutils. it completed more quickly when I changed the min object size from 5 pixels to 15. the time dropped from 50 to ~20 seconds, so that's reasonable.

[rfinn]

I tested on image pointing-13_R.coadd.fits from virgo-coadds-2017. The program is called get_galaxy_size.py.

INPUT

getobjectsize(image,xobj,yobj,scale=1.75,plotflag=False,usese=False)

RETURNS

return objsize

IMPLEMENTATION

need to integrate into halphamain.py

[rfinn]

Integrating into halphamain

• It's best to run this on the entire list of galaxies, rather than one galaxy at a time, because the biggest time overhead is running the detection algorithm.
• We can do this once per pointing, just using the r-band image
• get_gal_list creates self.gximage and self.gyimage. It's best to run getobjectsize after this step.
• get_gal_list is called by find_galaxies. Can insert getobjectsize right after find_galaxies call. Need to pass in r-band image name **self.rcoadd_fname

Change any calls for cutout_size

• previous code uses self.cutout_size for the size of one galaxy in arcsec

• need to set to value for that galaxy

  self.cutout_size = self.cutout_sizes[self.igal]*self.pixelscale

• changed get_galaxy_cutout to use self.cutout_size

• self.cutout_size is a 2D array with the x,y dimensions. this is what Cutout2D wants. I think the slider reset doesn't do this correctly.

• in get_galaxy_cutout

      size = self.cutout_sizes[self.igal]
      self.cutout_size = u.Quantity((size, size), u.arcsec)

  But according to the documentation, Cutout2D can take an integer or a 2D array. And it can take pixels or arcsec. So why do I convert to arcsec?

[rfinn]

Testing on Virgo INT 2017 data - p013

• worked out some more issues with cutout size. I had used it in pixels in some places and arcsec in others. So I made separate variables: cutout_size (pixels), cutout_size_arcsec (in arcsec, with unit attached)

[rfinn]

Testing on 2019, p0149

Problem with large galaxy. Box is crazy big

going to see how source extractor does with this.

The answer - not good.

[rfinn]

• I went back to using photutils. It's slow, but hopefully you only have to do it once.
• I have a max/min size set - this will catch any examples that are too big/small
• I am going to close this issue. We still should save the information somewhere. It is stored in bounding box in the output table. But maybe we want it somewhere else as well?