2010.02.09 15:03 "[Tiff] Comparing images to detect transmission errors", by Richard Nolde

Deborah,

There is a standard utility in the tools that come with libtiff called tiffcmp.

However, the conversion to fax may modify the image resolution depending on your faxing system. Standard faxes are 96 lines per inch and Fine Faxes are 196 lines per inch. (Or something close, I don't have the spec handy.) There are now much higher resolution options but you would need to know the correct values for the original and transmitted versions if you want to do a direct compare of the actual data.

 From the tiffcmp man page:

NAME
        tiffcmp - compare two TIFF files

SYNOPSIS
        tiffcmp [ options ] file1.tif file2.tif

DESCRIPTION
        Tiffcmp compares the tags and data in two files created
according to the Tagged Image File Format, Revision 6.0. The
        schemes used for compressing data in each file are immaterial
when data are compared-data are compared on a scanline-
        by-scanline basis after decompression. Most directory tags
are checked; notable exceptions are: GrayResponseCurve,
        ColorResponseCurve, and ColorMap tags. Data will not be
compared if any of the BitsPerSample, SamplesPerPixel, or
        ImageWidth values are not equal. By default, tiffcmp will
terminate if it encounters any difference.

OPTIONS
        -l List each byte of image data that differs between the files.

        -z number
               List specified number of image data bytes that differs
between the files.

        -t     Ignore any differences in directory tags.

BUGS
        Tags that are not recognized by the library are not compared;
they may also generate spurious diagnostics.

        The image data of tiled files is not compared, since the
TIFFReadScanline() function is used. An error will be
        reported for tiled files.

        The pixel and/or sample number reported in differences may be
off in some exotic cases.

Richard

I guess there is no paper involved?

If that's the case a very simple solution would be just so subtract the two images and count the non-zeros. You could normalise to the image size and calculate an error rate.