2000.11.14 00:58 "Re: Outstanding Libtiff Issues", by Tom Lane

http://bugzilla.remotesensing.org/show_bug.cgi?id=23

http://www.ai.sri.com/cgi-bin/digitalearth/list.pl?dir=sfbay.oi

The images at the URL above, in varying resolutions, are all tiled YCbCr TIFFs with JPEG compression. Not all tools handle YCbCr TIFFs or tiled TIFFs (or both), so it would be nice if tiffcp did. But despite its lack of diagnostics, when run on such an image with only the -s option, tiffcp produces a tiny file with mangled Y, Cb and Cr components (at least judging by XV's output). For example, the 512x640 level-4 image shrinks from 240k to 33k and changes from a reasonable aerial shot (with black regions indicating no data) to a mostly teal image.

FWIW, I can't duplicate this problem using my current copy of libtiff (CVS sources from mid-April 2000, not too current I fear). The downloaded images look a little funny in XV --- are the strips supposed to be out of registration like that? --- but they look just about the same after being run through tiffcp -s. They do get a lot smaller, which is not too surprising given that the originals were saved at Q 100 :-(... still another person who didn't read the libjpeg documentation I guess.

Since the tiles are (reported to be) 128x128 pixels, it should be possible to losslessly convert such an image into strips of 512x128 pixels (or so I would imagine--I'm no TIFF expert).

Lossless reformatting of TIFF/JPEGs would be a major project; you'd essentially need to rewrite tiffcp from scratch to make it deal with DCT coefficient blocks instead of pixels.

regards, tom lane
organizer, Independent JPEG Group