A high level utility to achieve decent quality image segmentation. It uses a mixture of image smoothing and watershed segmentation propagation to identify distinct objects for use in, e.g., profitSetupData (where the segim list item output of profoundMakeSegim would be passed to the segim input of profitSetupData).
profoundMakeSegim(image, mask, objects, skycut = 1, pixcut = 3, tolerance = 4, ext = 2,
sigma = 1, smooth = TRUE, SBlim, magzero = 0, gain = NULL, pixscale = 1, sky, skyRMS,
header, verbose = FALSE, plot = FALSE, stats = TRUE, rotstats = FALSE,
boundstats = FALSE, offset = 1, sortcol = "segID", decreasing = FALSE, ...)Numeric matrix; required, the image we want to analyse. Note, image NAs are treated as masked pixels.
Boolean matrix; optional, parts of the image to mask out (i.e. ignore), where 1 means mask out and 0 means use for analysis. If provided, this matrix *must* be the same dimensions as image.
Boolean matrix; optional, object mask where 1 is object and 0 is sky. If provided, this matrix *must* be the same dimensions as image.
Numeric scalar; the lowest threshold to make on the image in units of the skyRMS. Passed to profoundMakeSegim.
Integer scalar; the number of pixels required to identify an object. Passed to profoundMakeSegim.
Numeric scalar; the minimum height of the object in the units of skyRMS between its highest point (seed) and the point where it contacts another object (checked for every contact pixel). If the height is smaller than the tolerance, the object will be combined with one of its neighbours, which is the highest. The range 1-5 offers decent results usually.
Numeric scalar; radius of the neighbourhood in pixels for the detection of neighbouring objects. Higher value smooths out small objects.
Numeric scalar; standard deviation of the blur used when smooth=TRUE.
Logical; should smoothing be done on the target image?
Numeric scalar; the mag/asec^2 surface brightness threshold to apply. This is always used in conjunction with skycut, so set skycut to be very large (e.g. Inf) if you want a pure surface brightness threshold for the segmentation. magzero and pixscale must also be present for this to be used.
Numeric scalar; the magnitude zero point. What this implies depends on the magnitude system being used (e.g. AB or Vega). If provided along with pixscale then the flux and surface brightness outputs will represent magnitudes and mag/asec^2.
Numeric scalar; the gain (in photo-electrons per ADU). This is only used to compute object shot-noise component of the flux error (else this is set to 0).
Numeric scalar; the pixel scale, where pixscale=asec/pix (e.g. 0.4 for SDSS). If set to 1 (default), then the output is in terms of pixels, otherwise it is in arcseconds. If provided along with magzero then the flux and surface brightness outputs will represent magnitudes and mag/asec^2.
User provided estimate of the absolute sky level. If this is not provided then it will be computed internally using profoundSkyEst. Can be a scalar (value uniformly applied to full sigma map) or a matrix matching the dimensions of image (allows values to vary per pixel). This will be subtracted off the image internally, so only provide this if the sky does need to be subtracted!
User provided estimate of the RMS of the sky. If this is not provided then it will be computed internally using profoundSkyEst. Can be a scalar (value uniformly applied to full sigma map) or a matrix matching the dimensions of image (allows values to vary per pixel).
Full FITS header in table or vector format. If this is provided then the segmentations statistics table will gain RAcen and Decen coordinate outputs. Legal table format headers are provided by the read.fitshdr function or the hdr list output of read.fits in the astro package; the hdr output of readFITS in the FITSio package or the header output of magcutoutWCS. Missing header keywords are printed out and other header option arguments are used in these cases. See magWCSxy2radec.
Logical; should verbose output be displayed to the user? Since big image can take a long time to run, you might want to monitor progress.
Logical; should a diagnostic plot be generated? This is useful when you only have a small number of sources (roughly a few hundred). With more than this it can start to take a long time to make the plot!
Logical; should statistics on the segmented objects be returned? If magzero and pixscale have been provided then some of the outputs are computed in terms of magnitude and mag/asec^2 rather than flux and flux/pix^2 (see Value).
Logical; if TRUE then the asymm, flux_reflect and mag_reflect are computed, else they are set to NA. This is because they are very expensive to compute compared to other photometric properties.
Logical; if TRUE then various pixel boundary statistics are computed (Nedge, Nsky, Nobject, Nborder, edge_frac, edge_excess and FlagBorder). If FALSE these return NA instead (saving computation time).
Integer scalar; the distance to offset when searching for nearby segments (used in profoundSegimStats).
Character; name of the output column that the returned segmentation statistics data.frame should be sorted by (the default is segID, i.e. segment order). See below for column names and contents.
Logical; if FALSE (default) the segmentation statistics data.frame will be sorted in increasing order, if TRUE the data.frame will be sorted in decreasing order.
Further arguments to be passed to magimage. Only relevant is plot=TRUE.
A list containing:
Integer matrix; the segmentation map matched pixel by pixel to image.
Logical matrix; the object map matched pixel by pixel to image. 1 means there is an object at this pixel, 0 means it is a sky pixel. Can be used as a mask in various other functions that require objects to be masked out.
The estimated sky level of the image.
The estimated sky RMS of the image.
If stats=TRUE this is a data.frame (see below), otherwise NULL.
The header provided, if missing this is NULL.
The surface brightness limit of detected objects (requires at least magzero to be provided and skycut>0, else NULL).
The original function call.
Segmentation ID, which can be matched against values in segim
Unique ID, which is fairly static and based on the xmax and ymax position
Flux weighted x centre
Flux weighted y centre
x position of maximum flux
y position of maximum flux
Flux weighted degrees Right Ascension centre (only present if a header is provided)
Flux weighted degrees Declination centre (only present if a header is provided)
Right Ascension of maximum flux (only present if a header is provided)
Declination of maximum flux (only present if a header is provided)
Radial offset between the cen and max definition of the centre (units of pixscale, so if pixscale represents the standard asec/pix this will be asec)
Total flux (calculated using image-sky) in ADUs
Total flux converted to mag using magzero
Fraction of flux in the brightest pixel
Number of brightest pixels containing 50% of the flux
Number of brightest pixels containing 90% of the flux
Total number of pixels in this segment, i.e. contains 100% of the flux
Approximate elliptical semi-major axis containing 50% of the flux (units of pixscale, so if pixscale represents the standard asec/pix this will be asec)
Approximate elliptical semi-major axis containing 90% of the flux (units of pixscale, so if pixscale represents the standard asec/pix this will be asec)
Approximate elliptical semi-major axis containing 100% of the flux (units of pixscale, so if pixscale represents the standard asec/pix this will be asec)
Mean surface brightness containing brightest 50% of the flux, calculated as flux*0.5/N50 (if pixscale has been set correctly then this column will represent mag/asec^2. Otherwise it will be mag/pix^2)
Mean surface brightness containing brightest 90% of the flux, calculated as flux*0.9/N90 (if pixscale has been set correctly then this column will represent mag/asec^2. Otherwise it will be mag/pix^2)
Mean surface brightness containing all of the flux, calculated as flux/N100 (if pixscale has been set correctly then this column will represent mag/asec^2. Otherwise it will be mag/pix^2)
Weighted standard deviation in x (always in units of pix)
Weighted standard deviation in y (always in units of pix)
Weighted covariance in xy (always in units of pix)
Weighted correlation in xy (always in units of pix)
Concentration, R50/R90
180 degree flux asymmetry (0-1, where 0 is perfect symmetry and 1 complete asymmetry)
Flux corrected for asymmetry by doubling the contribution of flux for asymmetric pixels (defined as no matching segment pixel found when the segment is rotated through 180 degrees)
flux_reflect converted to mag using magzero
Weighted standard deviation along the major axis, i.e. the semi-major first moment, so ~2 times this would be a typical major axis Kron radius (always in units of pix)
Weighted standard deviation along the minor axis, i.e. the semi-minor first moment, so ~2 times this would be a typical minor axis Kron radius (always in units of pix)
Axial ratio as given by min/maj
Orientation of the semi-major axis in degrees. This has the convention that 0= | (vertical), 45= \, 90= - (horizontal), 135= /, 180= | (vertical)
Approximate singificance of the detection using the Chi-Square distribution
Approximate false-positive significance limit below which one such source might appear spuriously on an image this large
Estimated total error in the flux for the segment
Estimated total error in the magnitude for the segment
Sky subtraction component of the flux error
Sky RMS component of the flux error
Object shot-noise component of the flux error (only if gain is provided)
Mean flux of the sky over all segment pixels
Total flux of the sky over all segment pixels
Mean value of the sky RMS over all segment pixels
Number of edge segment pixels that make up the outer edge of the segment
Number of edge segment pixels that are touching sky
Number of edge segment pixels that are touching another object segment
Number of edge segment pixels that are touching the image border
Number of edge segment pixels that are touching a masked pixel (note NAs in image are also treated as masked pixels)
Fraction of edge segment pixels that are touching the sky i.e. NskyNedge, higher generally meaning more robust segmentation statistics
Ratio of the number of edge pixels to the expected number given the elliptical geometry measurements of the segment. If this is larger than 1 then it is a sign that the segment geometry is irregular, and is likely a flag for compromised photometry
A binary flag telling the user which image borders the segment touches. The bottom of the image is flagged 1, left=2, top=4 and right=8. A summed combination of these flags indicate the segment is in a corner touching two borders: bottom-left=3, top-left=6, top-right=12, bottom-right=9.
To use this function you will need to have EBImage installed. Since this can be a bit cumbersome on some platforms (given its dependencies) this is only listed as a suggested package. You can have a go at installing it by running:
> source("http://bioconductor.org/biocLite.R")
> biocLite("EBImage")
Linux users might also need to install some non-standard graphics libraries (depending on your install). If you do not have them already, you should look to install **jpeg** and **tiff** libraries (these are apparently technically not entirely free, hence not coming by default on some strictly open source Linux variants).
The profoundMakeSegim function offers a high level internal to R interface for making quick segmentation maps. The defaults should work reasonably well on modern survey data (see Examples), but should the solution not be ideal try modifying these parameters (in order of impact priority): skycut, pixcut, tolerance, sigma, ext.
See ?EBImage::watershed
profoundMakeSegimExpand, profoundProFound, profoundSegimStats, profoundSegimPlot
# NOT RUN {
image=readFITS(system.file("extdata", 'VIKING/mystery_VIKING_Z.fits',
package="ProFound"))$imDat
segim=profoundMakeSegim(image, plot=TRUE)
#Providing a mask entirely removes regions of the image for segmentation:
mask=matrix(0,dim(image)[1],dim(image)[2])
mask[1:80,]=1
profoundMakeSegim(image, mask=mask, plot=TRUE)
#Providing a previously created object map can sometimes help with detection (not here):
profoundMakeSegim(image, mask=mask, object=segim$objects, plot=TRUE)
# }
Run the code above in your browser using DataLab