The Center for Imaging Science has been provided with this series of one-foot IPR synthetic aperture radar (SAR) images courtesy of Thomas J. Burns, the MSTAR program manager for DARPA. This data was collected using the Sandia National Laboratories Twin Otter SAR sensor payload operating at X band. The data is collected and distributed under the DARPA moving and stationary target recognition (MSTAR) program.

There are a total of seventeen image sets, with each set containing data for a particular target imaged at a particular depression angle. Each set contains data for hundreds of degrees of target aspect pose for that target at that angle of depression.

?????.NXxNY.raw

The original data, containing an NX by NY series of floating point data representing the magnitude data, and another NX by NY series of floating point data representing the phase data.

?????.hdr.txt

A text file containing information about the target, the sensor, and the environment.

?????.gif

A GIF created by applying the following algorithm to the original data file:

For each pixel i
      mag  = magnitude[i]
      phs  = phase[i]
      real = mag * cos(2*PI*phs/4096)
      imag = mag * sin(2*PI*phs/4096)
    
    GIF[i] = log10(real*real + imag*imag)
EndFor

The MSTAR data files on the original CDs from SDMS begins with an ASCII text header describing the target, its pose, the data collection site, etc. Following this header is the raw, complex-valued SAR image data. This image data consists of a N_rows x N_columns array of pixel magnitudes (32 bit floating point) followed by an array of the same size containing phase data (in radians) for each pixel. The data is stored row-wise.

The MSTAR files available from the CIS web site have had the ASCII header removed. The following snippet of Matlab code will read a single image provided the file has been opened resulting in a file pointer, fp:

mstar_size = n_rows * n_cols;
[tmp_data, num] =
fread(fp,mstar_size*2,'float');

% Reshape works by column, but MSTAR data is stored by rows so the reshape
% arguments look funny.

tmp_mag = reshape(tmp_data(1:mstar_size),n_cols,n_rows).' ; 

tmp_phs = reshape(tmp_data(mstar_size+1:2*mstar_size),n_cols,n_rows).' ;

It is not clear how the GIF images were created but they were most likely taken off the CDs from SDMS. However, the above code will give you the floating point magnitude data which can be converted to gray scale with any histogram you like. I find a logarithmic transformation produces images that are pleasant to look at (you'll need to add a small constant, say 0.00001, to the magnitude data since some pixels may be zero). Of course, for ATR or other processing we use the magnitude data directly.

The values in tmp_phs are all in the range from 0 to 2pi if the data is read in correctly.

For further information, please contact Michael De Vore at http://cis.wustl.edu/~mdd2

This data is available with registered usernames and passwords. If you have already registered, and your registration has been confirmed, you may proceed to the download page. Where the following data sets will be available....