Given a target image, a segmentation map, image segstats and group properties, this function will attempt a non-parametric deblend based on local fitting of B-splines to create a weight map for each segment in a group. Flux is guaranteed to be conserved, and errors are appropriately rescaled.
profoundFluxDeblend(image = NULL, segim = NULL, segstats = NULL, groupim = NULL,
groupsegID = NULL, sky = 0, profound = NULL, magzero = 0, df = 3, radtrunc = 2,
iterative = FALSE, doallstats = TRUE, lowmemory = FALSE, deblendtype = 'fit', psf = NULL,
fluxweight = 'sum', convtype = 'brute', convmode = 'extended', Ndeblendlim = Inf)Numeric matrix; required, the image we want to analyse. As a convenience you can supply the output of profoundProFound of class profound, in which case any required input that is not explicitly set via the arguments will be inherited from the profoundProFound list.
Integer matrix; a specified segmentation map of the image. This matrix *must* be the same dimensions as image if supplied.
Data.frame, segmentation catalogue output from profoundProFound.
Integer matrix; the grouped segmentation map. This matrix *must* be the same dimensions as image. If missing then this is computed using profoundSegimGroup using the supplied segim.
List; group information as output by profoundSegimGroup or profoundProFound. Must correspond to the supplied segstats. If missing then this is computed using profoundSegimGroup using the supplied segim.
Numeric; the absolute sky level. Can be a scalar or a matrix matching the dimensions of image (allows values to vary per pixel).
List; object of class 'profound'. If this is provided then missing input arguments are taking directly from this structure (see Examples). As an added convenience, you can assign the profound object directly to the image input.
Numeric scalar; the magnitude zero point.
Integer scalar; degrees of freedom for the non-parametric spline fitting. See smooth.spline.
Numeric scalar; the maximum allowed radius beyond the edge-most segment pixel to consider when deblending. Keeping this low (1-3) ensures segments do not gather flux from very distant regions of the group.
Logical; should each segment profile fit be subtracted as it goes along? TRUE tends to remove the pedestal from a large galaxy that has faint objects embedded on top.
Logical; specifies whether the output catalogue is matched against all rows of the supplied segstats (TRUE), or only the rows containing grouped (and therefore deblended) galaxies are returned and the core flux columns (see below)s.
Logical; if TRUE then a low memory mode of FluxDeblend will be used. This is quite a bit slower, so the default of FALSE is usually preferred unless you are running into serious memory issues.
Scalar character, either 'fit' (where the segments in the image are approximately fitted for deblending) or 'psf' (where a PSF must be provided to argument psf). The former makes more sense for well resolved images (say optical) where the main issue in the deblending is the overlapping of resolved flux components. The latter works better in the regime where the image is barely resolved beyond the PSF, and this dominate the uncertainty of the deblend.
Numeric matrix; must be provided if deblendtype='psf'. This should be a small image (usually square) of the image PSF/beam to use for deblending.
Numeric scalar; only relevent when deblendtype='psf'. Either 'sum' (where the sum of the current segment weights the deblend solution), 'peak' (where only the peak pixel flux in the segment it used) or 'none' (no additional weighting is used). For very well resolved images 'sum' makes more sense, for barely resolved images 'peak' should be the safer option. Basically, you should pick the option that likely correlates best with the true (deblended) flux. This requires some thought on the user side (sorry!), but it might be a good idea to try both options and check the deblend quality.
Scalar character, only relevent when deblendtype='psf'. Specifies the type on convolver to use. Available options are displayed by running profitAvailableConvolvers, but usually one of 'brute' or 'fftw'. The latter tends to be fastest when the supplied psf is much smaller than the image (factor greater than 4 in each dimension). As the sizes become comparable, fftw becomes faster since it scales better with the product of psf and image pixels.
Scalar character, only relevent when deblendtype='psf'. Either 'extended' (the whole segment is convolved with the target PSF when determining the deblend) or 'psf' (the provided PSF is centred at the peak flux for deblending). Only relevent when deblendtype='psf'. The former makes more sense for well resolved images (say optical) where the main issue in the deblending is the overlapping of resolved flux components. The latter works better in the regime where the image is barely resolved beyond the PSF, and this dominate the uncertainty of the deblend.
Integer scalar; the limit for the number of pixels to consider in a deblending complex (Ngroup [number of segments in the group] x Npix [number of pixels in the group]). You might want to set this to a value similar to allowable machine memory (1e8 - 1e9 often) just to avoid extreme cases (e.g. large stars with lots of pixels and lots of segments).
A data.frame containing deblended flux information:
The group ID reference for the deblend (as taken from groupsegID)
The segment ID reference for the deblend (as taken from groupsegID)
Total flux (calculated using image-sky) in ADUs
Total flux converted to mag using magzero
Total number of pixels in this segment, i.e. contains 100% of the flux
Fraction of group flux in this segment. If this is very low then it is likely harder to extract good quality fluxes (with or without deblending).
Quality flag for deblended segment. This represents what fraction of the segment image flux is in our deblend model. Negative means the model misses flux, postive means it has too much. Nearer to 0 is better.
Quality flag for deblended group. This represents what fraction of the group image flux is in our deblend model. Negative means the model misses flux, postive means it has too much. Nearer to 0 is better.
Maximal beam correction. This is not applied, but for isolated sources this is the amount you should multiply the flux (and mag etc) given the beam PSF. In crowded fields this will not work well since flux will not be conserved globally!
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)
This routine only deblends with detected groups, so it is quite fast if the number of groups is quite low. If the image is more confused then this process can be quite slow.
Since there are a few ways to run it, here is some advice:
For clearly extended sources you probably want to run with deblendtype='fit', iterative=TRUE / deblendtype='psf', fluxweight='sum', convtype='extend'. The former works better for very complex source geometry (since it fits it), the latter might be a bit more stable though.
For poorly resolved images you probably want to run with deblendtype='psf', fluxweight='peak', convtype='psf'.
# NOT RUN {
image=readFITS(system.file("extdata", 'VIKING/mystery_VIKING_Z.fits', package="ProFound"))
profound=profoundProFound(image, magzero=30, verbose=TRUE, plot=TRUE, groupstats=TRUE)
deblend=profoundFluxDeblend(profound)
magplot(profound$segstats$mag, profound$segstats$mag-deblend$mag_db, ylim=c(-0.3,0.3),
grid=TRUE, xlab='mag', ylab='mag_orig - mag_deblend')
# }
Run the code above in your browser using DataLab