Bogus Huffman table definition

[ajan_p]

Hi,
I tried converting a DICOM image.
The details of the DICOM Image is as below
The transfersyntax is EXS_JPEGProcess14TransferSyntax
the dataset->chooseRepresentation(EXS_LittleEndianExplicit, NULL);
throws an error string "Bogus Huffman table definition". can any one help me sorting out this issue

[J. Riesmeier]

Have you already checked the source code of "dcmjpeg" for this kind of error code?

[ajan_p]

The issue was sorted out by a change in the source code of dcmjpeg library.

[J. Riesmeier]

So, does that mean that there is/was a bug in the dcmjpeg module, which is based on the IJG library, or is the image buggy?

[ajan_p]

The error was on IJG library.

[J. Riesmeier]

So, could you please tell us what was wrong and what you changed, i.e. could you provide a patch, so we can fix this bug also in the DCMTK. And, are you really sure that this issue was not caused by the DICOM image you've processed?

[ajan_p]

Please refer the site

https://gforge.sci.utah.edu/gf/project/ ... rowse&path

[J. Riesmeier]

Thank you. I guess you are referring to the following patch, right?

Code: Select all

Index: jdhuff.c
===================================================================
RCS file: /cvs/public/gdcm/src/jpeg/libijg8/jdhuff.c,v
retrieving revision 1.1
diff -u -3 -p -r1.1 jdhuff.c
--- jdhuff.c    24 Jun 2003 10:09:48 -0000      1.1
+++ jdhuff.c    7 Oct 2004 02:29:43 -0000
@@ -259,8 +259,11 @@ jpeg_make_d_derived_tbl (j_decompress_pt
   if (isDC) {
     for (i = 0; i < numsymbols; i++) {
       int sym = htbl->huffval[i];
-      if (sym < 0 || sym > 15)
-       ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+      if (sym < 0 || sym > 15) {
+         // Now, we can read Philips MRI Images
+         htbl->huffval[i]=15;
+//     ERREXIT(cinfo, JERR_BAD_HUFF_TABLE);
+      }
     }
   }
 }

If I interpret the comment correctly, Philips is doing something wrong with their MRI images, don't you agree?

[ajan_p]

absolutely right.

[Uli Schlachter]

Could you upload an example of such a broken image or could you send it to dicom (at) offis (dot) de?

[ajan_p]

uploaded the image to the email address

[Uli Schlachter]

Hi,

thanks for the image. However, I don't think we should apply your change to the ijg jpeg library.

JPEG splits the image up in blocks. For each block, the difference between the average luminance of this and the previous blocks gets entropy encoded. This is where the DC Huffman table that we have a problem with is used. Obviously, there is a maximum possible value for this difference. Your JPEG specifies how to encode a difference larger than this maximum possible value in its huffmann table.
Your patch only works because this image doesn't actually use the large value from the DC table.

I have no idea what would happen if this value actually gets used, but I guess that things would blow up in strange ways.

Also, I haven't found anything else which allows reading this broken image.

Uli