Opened 10 years ago

Closed 9 years ago

Last modified 9 years ago

#1882 closed defect (fixed)

Geotiff photometric minimum is white

Reported by: tomvr Owned by: warmerdam
Priority: normal Milestone: 1.5.0
Component: GDAL_Raster Version: 1.3.2
Severity: normal Keywords: gtiff


I tried to open a GEOTIFF file where the TIFFTAG_PHOTOMETRIC has been set to PHOTOMETRIC_MINISWHITE. It causes a segmentation fault (floating point exception). The problem occurs on line 2894 of the frmts/gtiff/geotiff.cpp file; here you try to build an inverse color pallette.

nColorCount = 1 << nBitsPerSample; nBitsPersample is 32 in my case ... for ( iColor = 0; iColor < nColorCount; iColor++ ) {

oEntry.c1 = oEntry.c2 = oEntry.c3 =

(255 * (nColorCount - 1 - iColor)) / (nColorCount-1);

oEntry.c4 = 255; poColorTable->SetColorEntry?( iColor, &oEntry );


'nColorCount' results to 1, resulting in a floating point exception since you are dividing by 0.

In attachment you can find the geotiff file I've used.

Attachments (1)

geotiff.tiff (556.8 KB) - added by tomvr 10 years ago.
test geotiff file

Download all attachments as: .zip

Change History (6)

Changed 10 years ago by tomvr

Attachment: geotiff.tiff added

test geotiff file

comment:1 Changed 10 years ago by Even Rouault

This bug still exists with GDAL trunk. I've commited in trunk in r12294 the following patch that checks that nBitsPerSample is 8 or 16 before creating a colormap. So in the case exhibited here, where the type is GDT_Float32, no colormap is created. But we don't handle the PHOTOMETRIC_MINISWHITE indication. I don't know how we could handle this situation without changing the values of the data...

Index: frmts/gtiff/geotiff.cpp
--- frmts/gtiff/geotiff.cpp     (révision 12286)
+++ frmts/gtiff/geotiff.cpp     (copie de travail)
@@ -2984,7 +2984,7 @@
        // Build inverted palette if we have inverted photometric.
        // Pixel values remains unchanged.
-       if( nPhotometric == PHOTOMETRIC_MINISWHITE )
+       if( (nBitsPerSample == 8 || nBitsPerSample == 16) && nPhotometric == PHOTOMETRIC_MINISWHITE )
            GDALColorEntry  oEntry;
            int             iColor, nColorCount;

Frank, do you think it should be backported to 1.4 branch ?

comment:2 Changed 10 years ago by Even Rouault

Smaller test case added to autotest in r12295

comment:3 Changed 9 years ago by Even Rouault

Milestone: 1.5.0

Should we consider the fix commited in r12294 as enough to close the bug ?

comment:4 Changed 9 years ago by warmerdam

Component: defaultGDAL_Raster
Keywords: gtiff added; geotiff removed
Resolution: fixed
Status: newclosed


I have modified the test a bit so that odd numbers of bits (4/12, etc) will also get color tables. With this minor adjustment I think things are fine. The situation is pretty esoteric, so I'm not going to back port into 1.4 branch.

comment:5 Changed 9 years ago by warmerdam

I neglected to mention my adjustment is r12777.

Note: See TracTickets for help on using tickets.