nproto.objmasks vs. xdimsum.makemask, imstat, etc.
Vallery Stanishev wrote on Oct 30, 2006
Hello,
I need advice/help on few questions:
1. Has anyone compared the performance of "nproto.objmasks" and "xdimsum.makemask" tasks? Any advice on which one (or anything else) to use when extended low-intensity sources are to be masked?
2. There are several options to compute image statistics using masks (proto.mimstat, xdimsum.miterstat,xdimsum.maskstat). Which on is to be preferred (faster, more reliable)?
3. Is there a task that projects a 2D image along given direction (by simply summing/averaging or by computing the median or mode) using object masks.
Thanks for the help.
/Vallery
I need advice/help on few questions:
1. Has anyone compared the performance of "nproto.objmasks" and "xdimsum.makemask" tasks? Any advice on which one (or anything else) to use when extended low-intensity sources are to be masked?
2. There are several options to compute image statistics using masks (proto.mimstat, xdimsum.miterstat,xdimsum.maskstat). Which on is to be preferred (faster, more reliable)?
3. Is there a task that projects a 2D image along given direction (by simply summing/averaging or by computing the median or mode) using object masks.
Thanks for the help.
/Vallery
Francisco Valdes wrote on Oct 30, 2006
Hello Vallery,
1. The two tasks you ask about have different algorithms. xdimsum.makemask is a single threshold method. This means the sky is determine and subtracted and then a single cut makes a mask of pixels above and below the threshold. proto.objmask does an (optional) convolution and then finds pixels above some number of sigma using a local sigma and sky. The pixels are combined together to define "objects" and each object is given a unique mask number. Most detection methods have difficulty with large extended sources. I would try both and see how it goes. Probably objmasks with a relatively large convolution would be more sensitive.
2. In general you should go with the core system tasks. The ones in xdimsum are scripts and ultimately call imstat. mimstat is a direct copy of imstat with the masks capability added. So use mimstat.
3. I can't think, off hand, of a task that will project the columns. But you can use imtranspose to orient an image (and the mask) so that the projection is along columns. Then I would use IMCOMBINE with the project=yes option. This means that it combines lines in a 2D image into a 1D image. If you really wanted you could then use imstat to promote the 1D image to 2D and then use imtranspose to make it a [1,*] image.
Yours,
Frank Valdes
1. The two tasks you ask about have different algorithms. xdimsum.makemask is a single threshold method. This means the sky is determine and subtracted and then a single cut makes a mask of pixels above and below the threshold. proto.objmask does an (optional) convolution and then finds pixels above some number of sigma using a local sigma and sky. The pixels are combined together to define "objects" and each object is given a unique mask number. Most detection methods have difficulty with large extended sources. I would try both and see how it goes. Probably objmasks with a relatively large convolution would be more sensitive.
2. In general you should go with the core system tasks. The ones in xdimsum are scripts and ultimately call imstat. mimstat is a direct copy of imstat with the masks capability added. So use mimstat.
3. I can't think, off hand, of a task that will project the columns. But you can use imtranspose to orient an image (and the mask) so that the projection is along columns. Then I would use IMCOMBINE with the project=yes option. This means that it combines lines in a 2D image into a 1D image. If you really wanted you could then use imstat to promote the 1D image to 2D and then use imtranspose to make it a [1,*] image.
Yours,
Frank Valdes
Last post on Oct 30, 2006