3dDiff


This is a program to examine element-wise differences between two images.

Usage

3dDiff  [display opt] [-tol TOLERANCE] [-mask MASK] <DSET_1> <DSET_2>

where:

  -tol TOLERANCE   :(opt) the floating-point tolerance/epsilon

  -mask MASK:      :(opt) the mask to use when comparing

  -a DSET_1        :(req) input dataset a

  -b DSET_2        :(req) input dataset b

... and there are the following (mutually exclusive) display options:

  -q               :(opt) quiet mode, indicate 0 for no differences and
                          1 for differences. -1 indicates that an error has
                          occurred (aka "Rick Mode").
  -tabular         :(opt) display only a table of differences, plus
                          a summary line (the same one as -brutalist)
                          Mostly for use with 4D data.
  -brutalist       :(opt) display one-liner. The first number indicates
                          whether there is a difference, the second number
                          indicates how many elements (3D) or volumes (4D)
                          were different, and the last number indicates the
                          total number of elements/volumes compared.
                          if there is a dataset dimension mismatch or an
                          error, then this will be a line of all -1s.
                          See examples below for sample output.
  -long_report      :(opt) print a large report with lots of information.

If no display options are used, a short message with a summary will print.

Examples

1) Basic Example: comparing two images
   A) In the 3D case, you get a short message indicating if there is no
      difference:
      $ 3dDiff -a image.nii -b image.nii
      ++ Images do NOT differ

      ... or a bit more information if there is a difference:
      $ 3dDiff -a mine.nii -b yours.nii
      ++ Images differ: 126976 of 126976 elements differ (100.00%)

   B) In the 4D case, the total number of elements AND total number of
      volumes which differ are reported:
      $ 3dDiff -a mine.nii -b yours.nii
      ++ Images differ: 10 of 10 volumes differ (100.00%) and 5965461 of 6082560 elements (98.07%)

2) A tolerance can be used to be more permissive of differences.  In this
   example, any voxel difference of 100 or less is considered equal:
   $ 3dDiff -tol 100 -a mine.nii -b yours.nii
   ++ Images differ: 234529 of 608256 elements differ (38.56%)

3) A mask can be used to limit which regions are being compared:
   $ 3dDiff -mask roi.nii -a mine.nii -b yours.nii
   ++ Images differ: 5 of 10 volumes differ (50.00%) and 675225 of 1350450 elements (50.00%)

   NB: The mask is assumed to have a single time point;  volumes in the mask
   beyond the [0]th are ignored.

Modes of output/reporting

There are a variety of reporting modes for 3dDiff, with varying levels
of verbosity. They can be used to view the image comparison in both human
and machine-readable formats. The default mode is the version shown in the
above examples, where a short statement is made summarizing the differences.
Reporting modes are mutually exclusive, but may be used with any of the
other program options without restriction.

1) Quiet Mode (-q)

Returns a single integer value in the range [-1, 1]:      -1 indicates a program error (e.g., grids do not match)
    0 indicates that the images have no differences
    1 indicates that the images have differences

Examples:
$ 3dDiff -q -a image.nii # no image b supplied
-1

$ 3dDiff -q -a image.nii -b image.nii # an image agrees with itself

$ 3dDiff -q -a mine.nii -b yours.nii # two different images

2) Tabular Mode (-tabular)

Prints out a table of values. Useful for 4D data, but not recommended
for 3D data.
Each row of the table will indicate the volume index and number of
differing elements.  At the end of the table, a summary line will
appear (see -brutalist).

Example (just using the first 10 volumes of two datasets):
$ 3dDiff -tabular -a "mine.nii[0..9]" -b "yours.nii[0..9]"
0:       596431
1:       596465
2:       596576
3:       596644
4:       596638
5:       596658
6:       596517
7:       596512
8:       596500
9:       596520
1 10 10 1.00000

3) Brutalist Mode (-brutalist)

   Creates a one-line summary of the differences. The numbers appear in the
   following order:
     Summary         [-1, 1], -1 failure, 1 differences, 0 agreement
     Differences     [0, NV/NT], the number of differing elements (3D) or
                     volumes (4D)
     Total Compared  NV/NT,      the number of elements/volumes compared
     Fraction Diff   [0, 1.0],   the fraction of differing elements/volumes

   Examples:
   $ 3dDiff -brutalist -a "mine.nii[0]" -b "yours.nii[0]" # 3D
   1 596431 608256 0.98056

   ... which means: There is a difference, 596431 elements differed,
   608256 elements were compared. The fraction of differing elements is
   0.98056.)

   $ 3dDiff -brutalist -a "mine.nii[0..9]" -b "yours.nii[0..9]" # 4D
   1 10 10 1.00000

   ... which means: There is a difference, 10 volumes differed, 10 volumes
   were compared.  The fraction of differing volumes is 1.0).

   If the program fails for some reason, brutalist output will be an array
   of all -1s, like this:
   $ 3dDiff -brutalist -a image.nii # no dataset b to compare to
   -1 -1 -1 -1

4) Long Report Mode (-long_report)
   Prints a very large report with lots of information.
   **WARNING:** this report is intended for use with humans, not machines!
   The author makes no guarantee of backwards compatibility for this mode,
   and will add or remove report outputs at his own (shocking whimsical)
   discretion.

Note on unhappy comparisons

If this program reports that the images cannot be element-wise compared,
you can examine the header information with 3dinfo. In particular, check out
the section, "Options requiring dataset pairing at input", most notably
options starting with "same", for example, -same_grid.

Author note:

Written by JB Teves, who notes:
  "Perfection is achieved not when there is no data left to
  add, but when there is no data left to throw away."