xmorph: Digital Image Warp ("morph") graphical user interface
Written and Copyright (C) 1994-1997 by Michael J. Gourlay
Provided as is. No warrantees, express or implied.
------------------
Modifications are often marked by a timestamp and the programmers' initials.
MJG refers to Michael J. Gourlay.
AT refers to Andy Thaler <thaller@informatik.tu-muenchen.de>
WA refers to Warwick Allison <warwick@cs.uq.oz.au>
MH refers to Mike Hoefelein <hoefelei@informatik.uni-wuerzburg.de>
PW refers to Paul Walker <pentode!paul@remlap.cts.com>
Thanks to Geoff Lawler for the Windows NT/95 cygwin port.
Other help from ellenberger@tle.enet.dec.com
Huge help from Dave Beazley, author of SWIG, for wrapper code
generation help (non-specific to tkmorph).
Gordon Matzigkeit <gord@gnu.ai.mit.edu> added the GIMP plugin
functionality.
------------------
20jul99:
added #include<stdlib.h> in my_malloc.h for when DEBUG is not defined
15jul99:
separated out libmorph as a stand-alone library.
wrote "morph" command line interface version.
19apr98:
GG added a missing meshInit in warp_sequence that was causing
memory problems on some machines.
17apr98:
Added red-green color-blindness changes for GUI
23sep97:
Added GIMP plugin functionality, enabled with -DGIMP in the Makefile.
19jul97:
No longer rebuild entire "wish" executable. Uses DLL and DSO instead.
Renamed "main.tk" to "tkmorph" and changed executable from "tkmorph"
to "wish4.2".
14jul97:
Received binary executable for xmorph on Windows NT/95 under cygwin.
09jul97:
Added meshSet and changed diw_map to use it.
Added meshCopy, meshStore, meshRecover, meshBackupIndexSet and other
associated "undo" routines and incorporated them into each mesh
modifying routine that a user would call during mesh editting.
08jul97:
Added meshLineMouseModify to mesh.c and changed diw_map.c to use it.
02jul97:
Removed mesh_diw.[ch] since it contained only a call to meshReset.
Incorporated use of meshScale when loading images of a size different
from the previous size.
early June 1997:
incorporated TCL/Tk and SWIG code into main xmorph code.
separate tkmorph code is now obsolete. New code requires SWIG 1.1b4 or later
18apr97:
Added Apollo Domain/OS changes, due to PW.
Changed the name of the my_malloc macros to remove the MY_ prefix.
06dec96:
Added -DREAL=double to Makefile for spline.c
24nov96:
Added meshInitialize to initialize new mesh members, intended
primarily to set the mesh arrays to NULL so that the revised
meshAlloc routine didn't try to free invalid memory addresses.
Added meshMatch to match geometries of two meshes,
intended for use in loading new meshes to make sure both meshes
are the same size, in a compact routine.
Changed meshAlloc to free old space before allocating new space.
This required changes elsewhere to make sure that previously
unallocated mesh arrays were explicitly set to NULL.
This could cause weird memory bugs later if I didn't catch all of
the meshAllocs.
13nov96:
Added a check to make sure that the start and finish images are the
same size in the case that they are provided as command line
arguments. Reported by Rick Walker <walker@cutter.hpl.hp.com>
02oct96:
Fixed the meshLoad routine to read in longs instead of ints for nx,ny.
Thanks to George C. Hawkins for pointing it out. The bug was
introduced in mid August when the mesh data structure was rewritten.
nx and ny used to be regular ints. Now they are longs. I forgot
why I bothered to change their type, and I wonder what other things
depend on them being regular ints instead of longs.
Fixed a bunch of tiny little warning-level bugs, such as including
needed include files, providing strong prototypes for callback
member in the mjg_dialog, fixing a compiler macro in one of the
mesh_.h files, etc.
Included a copy of the tkmorph stuff so that people can see what I'm
up to, in case they want to contribute or something. However, it's
not production code.
10sep96:
fixed init_diw_stuff in diw_map.c where color allocation used to
only be done for PseudoColor. It is now also done for all visual
classes. The effect was that the mesh lines were being drawn in
black for TrueColor visuals.
30aug96:
Created mesh_diw.[ch] and moved reset_meshes routine from mesh.[ch]
to mesh_diw.[ch]. Renamed many of the mesh routine names to use
another function name convention.
Rewrote the entire mesh part of the code to use a "meshT" user
defined data type. Many dozens of lines of code throughout many
modules were modified. I should have done this to begin with, but
I didn't originally, because I wanted to have the source and dst
meshes have identical number of points, which I decided was best by
just having a single member for the size of both meshes. However,
the TCL version will be easier to handle, and the code is more
elegant, with a "meshT" data type. This more than likely
introduced bugs into the code, particularly regarding any code that
modifies or uses the "number of points" members, and in particular,
of the "dst" meshes.
09aug96:
image.c and image.h renamed to rgba_image.[ch]. The routines in
rgba_image.c were renamed to use a different routine naming
convention. For example, load_img was renamed loadImageRGBA, etc.
26jul96:
Moved reset_images, orig_image, and dither_image into the new
module image_diw because they are special purpose code used only by
xmorph, whereas the other image.c code is general purpose.
12jul96:
Fixed the dithering routine to make use of the dithiering error
accumulator. Looks better now. The images used to dither sort of
okay before, but it was really stupid of me not to expect better
results out of this dithering technique. I guess I figured that
the technique wasn't any better than the results i was getting
before. Ends up, it was just buggy code.
Added support for StaticGray visual class with 1-bit depth.
To do this, I had to add a section to the dither_image routine to
handle StaticGray. The dithering routine I cooked up for the color
images didn't translate well for only 1-bit because the error
diffusion of the color scheme is very unsophisticated because it
relies on the assumption that there won't be enough quantization at
any given pixel to make a harsh error. Well, with 1-bit depth, the
quantization error is almost always drastic, so I had to implement
something more like a full-blown Floyd Steinberg error diffusion
algorithm. Mine is still simpler than full-blown FS, but it looks
about as good on a flying horse.
28jun96:
Changed the way colormaps are allocated so that now it is a trivial
matter to change the number of bits ber channel used to represent a
color in the dithering colormap. The primary changes were in
diw_map.c and diw_map.h, although some changes were in image.c in
dither_image. Note that there are some limitations on the maximum
number of bits per channel useable due to the BITS_PER_CHANNEL in
dither_image, so 16-bit visuals might not be able to make full use
of their huge colormaps. Since I don't have such a machine, I
wouldn't be able to test it easily. It might be as simple as
changing BITS_PER_CHANNEL to 16 and cranking up the ???_NUM_BITS
values to something like 3 or 4 bits each (perhaps RGB-443). This
would yeild VERY nice looking images (almost at TrueColor quality)
and the dithering would probably be just as fast as it is now. I'm
stoked.
30may96: MH
Fixed a subtle but important bug in warp.c line 189:
/* scan convert vertical splines of S and I */
hermite3_array(yrow1, xrow1, mesh_height, indx, map1, img_height);
hermite3_array(yrow1, xrow2, mesh_height, indx, map2, img_height);
^^^^^
The report comes from Mike Hoefelein
<hoefelei@informatik.uni-wuerzburg.de>
The effect was to make the warped image conform exactly to the
mesh. Before, there was a descrepency between the mesh location
and the image. This should have been obvious earlier by simply
warping the checkerboard test image, but I had ignored it until
Mike pointed it out to me.
14nov95:
Fixed a hideous memory allocation bug.
TrueColor works for TrueColor default visual.
TrueColor does not work for nondefault TrueColor, because Xt does
not support such a use of Visual Class. This ends up being a
rather subtle and difficult issue, which has been discussed at
length by people who want to use TrueColor visuals on displays
that have another kind of default visual class but which can
support TrueColor. See the X Authority Web site.
14sep95:
Removed TrueColor support for nondefault TrueColor machines, but
left hooks in. I can't figure out why it doesn't work.
13sep95:
Made some modifications due to reports by xmorph users.
In particular, there were more divide-by-zero events happening in
the warping code. Much of this was due to Wolberg. I've found
several errors in Wolberg's original code, and he's been made aware
of these changes. One good thing about these changes is that they
(in Wolberg's opinion) sufficiently change the code such that there
isn't a copyright problem using the code. Wolberg has granted
permission to me to use the code in any case.
Added my_free to spl-array in hermite3_array (memory leak).
Changed free() to my_free in various routines.
Changed xi loop bound in reample_array in warp.c to (len-1)
WA contributed these changes:
added TrueColor support (changed xbm_gray and dither_image)
I have to admit that I don't fully understand these changes yet
because I haven't looked into TrueColor visuals because I
haven't hacked them. If there is a problem with the TrueColor
code then let me know, but I might not be able to help. -- MJG
fixed memory leak in sequence.c to free a temporary mesh.
changed box to paned widget to allow scrolling
14jul94:
Modified spline routines to fit my personal library better.
Moved some routines from spline.c to spl-array.c
Miniscule formatting changes in my_malloc.c.
Nov94:
Changed my email address to Colorado address in documentation files.
Released at MIT.
22jul94:
More diddling with the tga routines. Colormap and pixel stuff.
21jul94:
removed load_tga_cmap_entry, and changed the way targa cmaps are stored.
Added ppmtotga, tgatoppm, cjpeg, djpeg to the man page.
release 5 (xmorph-20jul94)
20jul94:
Fixed load_tga_image: first row of upside-down images was wrong.
Removed (or changed) references to tga.h where inappopriate.
Added automatic image compression in save_img if loaded image was
compressed. Also checks pixel_size.
Added fields to the rgba_image_t for compression, cmaps, pixel size.
Added RLE compression to Targa save routines.
Added a bounds check in load_tga_image for corrupt Targas.
Added more labels to the scrollbars to be more clear.
Alligned scrollbar region.
Added a missing "#include <Scrollbar.h>" in xmorph.c
Added a missing "\n\" in help_menu.c
Alpha version: these releases come all too often.
19jul94:
Ximage array allocation fixed in diw_map.c (factor of 4 deal).
Fourth release (xmorph-19jul94.2): unannounced
Ximage array indexing fixed in dither_image.
Third release (xmorph-19jul94.1): unannounced
Modified spline code to continue (not abort) when spline abcissa out of range.
Second release (xmorph-18jul94): unannounced
Tue Jul 19 02:57:37 EDT 1994
This release was not quite ready, but I was being flooded with
reports that xdiw had vanished, and I needed to put something up to
keep the email flux down.
xmorph-18jul94:
Removed spline code from warp.c.
Incorporated use of general purpose hermite spline routines I wrote in 1993.
Fixed bug in hermite spline routine (for point at last knot).
Changed the name from xdiw to xmorph.
xdiw-15jul94:
Changed the module organization of the rgba_image_t definition.
Separated tga-24.c and (new file) tga-write.c
Added size-8 write to save_tga_pixel (untested)
[Approaching a common tga library for xdiw and showtime]
xdiw-08jul94:
Added authorship and release date info in internal "About..." page.
Made README more like what the X Consortium wants for contrib.
Separated application and widget code in diw_map.c and xdiw.c.
Wrote a short man page for xdiw.
First release (xdiw-07jul94): announced
Thu Jul 7 16:36:58 EDT 1994:
It's at a stage I'd call "alpha release" if I were into that sort of
jargon, which means this program will probably dump core a lot.
Please report such incidents, along with any solutions, back to me.
I've used it to morph images of Wendy and myself, with significant
success. I put the images onto video and it looks really nice.
I've never seen any other morphing program at all, so I'm sure my idea
of a GUI differs greatly from the "professional" commercial packages,
but according to one guy I showed this to, my way is better. If you
disagree, please tell me what I can do to make my program more usable.
I already have a lot of ideas about how to make it better-- features
to add, etc., but at the moment it's pretty usable (enough so that
I've decided to release it.)
It's self documenting, and I've compiled and run the program on SunOS
machines, Solaris machines, SGI's, and HPUX, although on the HP it's
slower than molasses. It _flies_ on the Reality Engine.
I'm trying to figure out how to get this widely distributed. As far
as I know, it's the only publically available image morphing program,
for X or any other platform. If any of you know of another publically
available morphing program, please let me know so I can kill myself.
I also wrote a program that rapidly displays the images generated by
Xdiw (for X or MSDOS), for those of you who don't have a high speed
JPEG CoDec like I do :)
