Package: stsdas.analysis.restore
Help file updated: Aug93
The restore package contains tasks to deconvolve or filter
1- or 2-dimensional images.
The lucy task generates a deconvolved image from an input image and
point spread function (PSF). The output image will be the same size as
the input image unless an optional larger output size is specified by the
user. The PSF image must contain the same number of dimensions as the
the input image, but the actual dimensions are not required to match the
input image. If a background (sky) image or mask image are given, their
dimensions must match those of the input image. If a model image is
specified its dimensions must match the output image size. When the
optional background image is specified the lucy task will require 6 real
and 2 complex arrays, i.e. a 512x512 image will require 10 MB of array space.
The mem task restores images by the Maximum Entrophy Method. The input
images are mainly (1) the degraded image (e.g., from the WF/PC of the HST),
(2) the point spread function image, and optionally (3) the model image.
The input parameters are the readout noise and adu (the A/D conversion
constant). The output image is the final converged MEM image, or it
may be an interim result which may be used as a model to run the task
again.
The sclean task implements the sigma-CLEAN deconvolution algorithm of
W. Keel (PASP 1991, 103,723). The task includes some additional features
not contained in Keel's code: (1) the optional convolution with the
"restored beam", (2) the optional addition of the residual map to
CLEAN map, and (3) a much faster algorithm.
The wiener task applies a Fourier noniterative deconvolution filter to
2-dimensional images. The filter includes the familiar inverse, Wiener,
geometric mean and parametric forms. Type "help wiener opt=sys" to see a
more detailed explanation on filter characteristics, limitations and
usage. You may use the task with its default parameter settings,
supplying only the required parameters. However, to fully exploit the task
(and hopefully achieve better restorations), reading
the opt=sys help pages is highly recommended.
The hfilter task implements a simple realization of an adaptive low-pass
noise filter based on the H-transform. The H-transform is a form of image
coding based on the Haar transform, which keeps a tight one-to-one
relationship between "pixels" in the transform space and data pixels
in the image space. With it, it is possible to design filters that
adapt themselves to the local signal-to-noise and spectral properties of
the data. A complete description of the technique can be found in the
references in the help page.
The adaptive task, written by G.M. Richter at ESO,
reduces the noise in portions of an image without reducing the resolution
of sharp features. The local signal-to-noise ratio as a function of
decreasing resolution is evaluated via the H-transform in the following
way. At a given point in the image, mean gradients and curvatures over
different scale lengths (obtained from the H-coefficients of different
orders) are compared to the corresponding expectation values of the
noise. The order for which this signal-to-noise ratio exceeds a specified
level indicates the local resolution scale length of the signal, and this
determines the size of the filter to apply at this point.