This directory contains a newer/customized version
of FR/trans as of 08/03/2001.

What's new:

- Code for iterative transmission reconstruction is more modular

- Simple ascii i/o instead of matlab i/o



Sections in this file:
  I.   Syntax of the trans command 
  II.  Changes in the code/functionality 
  III. Examples



I. The syntax of the trans command:
---------------------------------------------------------------------

Usage: trans [-x {#pix_x} (n_t)] [-y {#pix_y} (n_t)] 
     [-f {field_of_view_x_in_mm} (n_t*delta_t)] [-i {#iter_recon} (20)] 
     [-E {#iter_param} (0)] [-s {sigma} <not with -E>] 
     [-p {GGMRF_shape_param} (2.0)]  [-I init_file] [-S support_file] 
     [-R pixelreplicate_result]  inputfile outputfile

Values in () indicate default settings.



inputfile:  File containing sinogram and sinogram parameters

outputfile: filename for output file.


OPTIONS:

-x #xpix :  x-dimension of reconstruction image (number of pixels)
            the default is the number of displacements n_t.

-y #ypix :  y-dimension of reconstruction image (number of pixels)
            the default is the number of displacements n_t.

-f {fovxmm} : reconstruction field of view in x-direction in mm. 
            This will implicitely also specify the fov size in y
            because the pixels are assumed to be square.
            default is size of projection array n_t*delta_t 

-i #iter_rec : number of reconstruction iterations
            default is 20

-E #iter_param: number of iterations for estimation of sigma
            default is 0 which means parameter estimation is
            switched off. In that case, a sigma must be
            specified with the -s flag.
            Note, that you cannot use both -s and -E flags at
            the same time.

-s {sigma} : Specifies the sigma parameter. Must be used when
            -E is not used or with -E0. Cannot be used with -E #iter_param
            when iter_param>0.

-p {GGMRF_shape_param} : specifies GGMRF shape parameter. Should
            be in interval (1, 2]. Default is 2.0 which is
            Gaussian MRF case.

-I init_file: Initialization file for reconstruction that must be
            of same geometry/pixel-size as reconstruction image.
            If this is not specified, the algorithm will initialize
            the image by a constant=0.1.
            This default initialization should only be used
            for VERY LOW RECONSTRUCTION RESOLUTION, such as 32 x 32
            pixels. It is much faster to then perform a multiscale
            initialization (see examples) as compared to initializing
            a high-resolution reconstruction with a constant.

-S support: specifies file containing mask of image support 
            This is only needed when parameter estimation is
            switched on, i.e. -E flag is specified.

-R :        interpolate output by a factor of 2 in x and y direction
            using pixel replication. This is practical for a multiscale
            initialization strategy where the current reconstruction is
            to be used to initialize follow-up reconstruction at 
            the next higher resolution. (see examples).


II. Changes in the code/functionality
-------------------------------------------------------------------

1) projection matrix is stored instead of computed on the fly.
   This speeds up the reconstruction for the old example by a
   factor of approx. 2.5. 
   If you don't have enough memory and you want to revert to 
   computing the proj. matrix on the fly, comment out the

   #define STORE_A_MATRIX

   in ../ansi_lib/A_comp.c

2) Code has been changed to use double arithmetic instead
   of floats. However, the projection matrix is still stored
   at floats in order to conserve memory.

3) Code allows for rectangular reconstruction geometry. The
   geometry can be specified using combinations of the 
   -x, -y and -f options. 
   NOTE: This CANNOT be used to perform region of interest 
         reconstructions!! The reconstruction image MUST
         contain all objects that were imaged by the projection
         measurements. If you leave objects outside the reconstruction
         field of view, the forward projection of your reconstruction
         image won't match the measured projection data and you will 
         get artifacts.

4) Added pixelreplication option to facilitate multiresolution
   initialization.



III. New examples
-----------------------------------------------------------------
1) EXAMPLE_OLD_EQUIV: This is the equivalent of the old 
   EXAMPLE_TRANS using the new command line syntax.

   This example requires approx. 123MB of RAM when the projection
   matrix is stored.

2) EXAMPLE_MSCAL_SQUARE: Performs same 256x256 reconstruction
   as example 1), but uses multiscale initialization instead
   of CBP initialization. That is, the EXAMPLE_MSCAL_SQUARE is
   a shell script that first calls trans to perform a low resolution
   reconstruction of 32 by 32 pixels. The pixel-replicated reconstruction
   is then used to initialize a 64 by 64 pixel reconstruction 
   and so forth until we reach 256 by 256. This is actually
   faster overall than using the CBP initialization. Specifically,
   after 20 iterations, the 256 by 256 reconstruction reaches the
   same convergence as the CBP initialized solution after 70 iterations.
   A difference to 1) is that the sigma is specified, not estimated.

   This example requires approx. 123MB of RAM when the projection
   matrix is stored.


3) EXAMPLE_MSCAL_RECT: Same as 2) but with rectangular reconstruction
   geometry. 128 pixels in y-direction and 256 pixels in x-direction.
   Field of view in x is default, i.e. n_t*delta_t.

   This requires approx. 70MB of RAM when the projection matrix is 
   stored.

4) EXAMPLE_MSCAL_RECT2: Same as 3) but smaller field of view.

 
  

