🚀 go-pugleaf

NEW! And Improved!


Version 0.2 of RLAB, a collection of a few routines I have written
is now available. There are several restoration routines and currently two 
registration routines. The focus of the restoration routines is primarily on the
technique commonly refered to as regularization.  

In going from 0.1 to 0.2 I have corrected a few bugs,
added two routines, and included 
workspaces that show that usage of each routine. I also have taken out
a routine or two that I didnt feel was useful.

Here is a listing of the
routines and some references if you want to learn more. I hope to have
a JPL external publication coming out soon which explains how each of
the algorithms work but, until then, you'll have to check the references.

Obtaining RLAB: RLAB can be obtained through anonymous ftp from the site
elroy.jpl.nasa.gov in /pub/mathew/rlab-0.2.tar.Z



RESTORATION

vmanual - manually lets the user step through a series of values which 
determines the degree of regularization (smoothing). The misfit values
can be saved for plotting with xprism2. The restored image, for each
regularization value, can be saved and viewed with Animate.

vchi - The regularization parameter is chosen so that the misfit satisfies
a chi squared constraint.

vbayes - the regularization parameter is selected that is most probable
given the data.

vpremodel - same as vbayes except the image prior is not assumed to be flat
but, instead, is the result of convolving a flat prior "pre"- image 
with a blurring kernel. This is Stephen Gulls "Intrinsic Correlation" model.


lucy - This is a Cantata workspace which implements the Lucy-Richardson
restoration algorithm


REGISTRATION

vrotmatch - template matching on rotated images. A set of rotationally
invariant coefficients is taken from the template image and compared
at different locations on the target image. (This rotuine was called
vmatch in 0.1)

vtranmatch - fast template matching on nonrotated images. I use an
iterative schemes that results in an algorithm with speed independent
of window size and depending only linearly on the size of the search
region.

RECOGNITION

vtexclass- texture classification using sum and difference histograms
vcohist - computes sum and difference histograms for use in vtexclass.


MISC

cipe2viff - converts a raw file to a viff file by using an auxillary
header file. The header file contains keyword value pairs specifying
thje values you normally would give to raw2viff. In fact, except for
the header reading code, this is just a trivial front end to raw2viff.


vmoments - Calculates a set of rotationally invariant coefficients 
about a given pixel in a circular window of desired radius. Optionally,
take samples from the window.

vmann - generates the Mann-Whitney distribution for a given choice of
size for each of two sample sets. Both this and vmoments are used in
vmatch.

vcfft2 - 2d non radix 2 fft's.

vcorr - spatial correlation. mask is centered over pixel so there is no offset
in the resulting image. Boundaries are handled  by extending the image.

vbandmin - in a multiband image, replace pixel value with bandnumber of
smallest value.

FOR FUN

I have included a hack of the Khoros package Warpimage which I call morphtool.
The idea is to create animation sequences like the kind you see in a lot
of commercials these days where one face slowly changes into another.
This is really a hack and included just for fun. 

REFERENCES - 

vchi,vbayes - check out Thompson et al , "A Study of Methods of choosing
the smoothing parameter in image restoration by regularization"
IEEE PAMI Vol 13, No 4, 1991.

vpremodel - Gull, "Developments in Maximum Entropy and Data Analysis" in
Maximum Entropy and Bayesian Methods, Cambridge England, 1988.



vmatch,vmoments,vmann - Goshtasby, "Template Matching in Rotated Images"
IEEE PAMI Vol 7, No 3 1985.

lucy - Lucy, "An iterative technique for the rectification of observed
distributions", The Astronomical Journal June 1974.

vtexclass, vcohist - " Sum and Difference Histograms for Texture 
Classification", IEEE PAMI vol 7 No 1 1986.


Mathew Yeates