Menu
▾
▴
[f4ea6f]: / NEWS Maximize Restore History
901 lines (581 with data), 29.9 kB
* Version 4.3
Unreleased. Current line of development.
** Improvements
- All deprecated options and features of version 4.2 have been
removed.
- Enblend and Enfuse transfer EXIF, IPTC, and XMP meta-data from the
input images to the output images.
- Both Enblend and Enfuse can now send their output to files without a
filename extension, which included devices, like e.g. "/dev/stdout".
The output format is PNM (Portable aNyMap). The same holds for the
output mask file. The common abbreviation `-' for `/dev/stdout' is
understood.
- Enfuse: The default saturation weight has been set to zero. This
makes Enfuse's behavior more predictable.
** New Commandline Options
- To circumnavigate the lack of alpha-channel support of some common
image formats, add option `--output-mask', to let the user define a
mask filename for the output.
** Developer Stuff
- Configuration switches `--enable-image-cache' and its opposite were
downgraded to nops.
- The new configuration option "--enable-partially-static-linking"
controls whether all libraries are linked in their shared versions
(default) or some performance critical libraries are linked in with
the static versions.
** Package Maintainer Stuff
- All Boost dependencies have been scrapped.
- The environment variable SOURCE_DATE_EPOCH overrides the build
timestamp as recorded by the signatures. See
https://reproducible-builds.org/specs/source-date-epoch/
for details.
* Version 4.2 "Compressor Road"
Released on March 29, 2016.
** Improvements
- All deprecated options and features of version 4.1 have been
removed.
- Enblend and Enfuse support layer (aka "page", aka "frame") selection
in of multi-layer image-files. This includes processing any layers
in user-defined order.
- Both Enblend and Enfuse can now perform all their pyramidal blending
operations in CIELAB (aka L*a*b*, aka L-star) and CIELUV (aka
L*u*v*) color spaces, too. See section "New Commandline Options",
option `--blend-colorspace'.
The previous default, CIECAM for input images with ICC-profile has
been changed to CIELUV. The default for RGB images without
ICC-profile remains. However, the new default for floating-point
image data is to use the RGB-cube, no matter whether the images come
with profiles or not.
- The new Graph-Cut algorithm is the default primary seam-line
generator.
- Avoid a division-by-zero in the Annealing Optimizer.
- Fix a bug (#1356551) in the seam-line vectorization code that was
there since 2004. The fix changes the position of almost any
coarse-mask seam line vertex by one pixel.
- Fix a longstanding quirk, which allowed to load masks into Enblend
and Enfuse that were unsuitable for processing.
- Fix a bug in the highlight-recovery that caused Enfuse to bail out
with the uncaught exception
"Minimizer1D::set_bracket: minimum not bracketed".
- The OpenMP enabled versions of Enblend and Enfuse benefit from
Google's TCMalloc library.
- Several exposure weight functions were added to Enfuse. The choice
is now up to the user. The default still is the Gauss function as
in all Enfuse versions before.
- Enblend and Enfuse now both can make use of the tremendous computing
power in modern graphics cards. To that end the binaries must be
compiled with OpenCL support and the necessary drivers and or
libraries must be installed on the host system.
- Enfuse can access the OpenCL compiler to allow for user-defined
exposure weighting functions specified in OpenCl (source) files.
** New Commandline Options
- The introduction of additional blend color spaces requires a new
option to select them. The new option `--blend-colorspace' takes
over the functionality of `--ciecam' and `--no-ciecam' (which both
will be withdrawn in a future version) and is flexible enough for
the new CIELAB and CIELUV blend color spaces.
--ciecam => --blend-colorspace=CIECAM
--no-ciecam => --blend-colorspace=RGBCUBE
new: --blend-colorspace=CIELAB
new: --blend-colorspace=CIELUV
- Calling Enblend or Enfuse with `--show-software-components' will
show the compiler name and version along will the versions of all
important libraries against which the binaries were compiled.
- The new options `--show-globbing-algorithms',
`--show-image-formats', and `--show-signature' now display the
information that previously was requested with `--version
--verbose'. The output of `--help' has been regrouped to account
for the new diversity of information options.
- Option `--exposure-weight-function' makes Enfuse's exposure weight
function user selectable. Previously weight function "gauss" was
the only one (and hard-coded).
Some examples of user-defined weight functions and a rudimentary
GNUmakefile to build them get installed in
DOCDIR/examples/enfuse.
- The old Enfuse options `--exposure-mu' and `--exposure-sigma' got
siblings called `--exposure-optimum' and `--exposure-width'
respectively. With exposure weight functions other than Gauss the
restricted naming does not make sense anymore.
Options `--exposure-mu' and `--exposure-sigma' are scheduled for
withdrawal in a later version of Enfuse.
- For a long time Enblend has had an option to run the seam-line
optimizer on the GPU. Now, both Enblend and Enfuse sport `--gpu'
options to offload computationally intensive tasks to any
OpenCL-compatible GPU device.
On systems with more than one device `--prefer-gpu' selects the
desired GPU and `--show-gpu-info' lists the available GPUs.
- Long-time option `-a' got a long-form sibling called
`--pre-assemble'. The long form can be negated as
`--no-pre-assemble', which is the default.
** Developer Stuff
- During configuration the name of Vigra's import-export library can
be set with the environment variable VIGRAIMPEX_LIB. This is to
splice in a library compiled with e.g. instrumentation or
sanitizers. If the variable is undefined or empty the default is
`-lvigraimpex', which lets most linkers prefer the dynamic library
in the standard location if there is any choice.
- The Automake system can be configured to print only terse messages
(configure option `--enable-silent-rules'), or when working at
command line level, pass `V=0' to make(1).
- A C++11-compatible compiler is required to compile Enblend and
Enfuse.
- Support of the Boost filesystem library has been dropped. There is
no Boost library dependence anymore.
- Empty `--parameter' options default to "1". This behavior in
conjunction with boolean parameter expansion taking "0" and "1" as
aliases for "false" and "true" allows to shorten setting boolean
parameters.
- The new configuration option "--enable-on-demand-dynlink" controls
lazy loading of all shared libraries.
** Package Maintainer Stuff
- Boost version 1.55 or later is now required to build.
- Vigra version 1.9 or later is now required to build.
- Little CMS version 2.5 or later is now required to build.
- Support for OpenGL acceleration has been dropped. So, the
dependencies on the libraries GL, GLU, and Glut do not exist any
longer.
- Support for images with 8-bit signed channels has been dropped.
- The documentation no longer uses Texinfo, but LaTeX for PostScript
generation and Hevea for HTML.
- fig2dev(1) of the XFig suite is no longer needed to build the
documentation. All relevant images were converted to dot(1) or SVG
format.
* Version 4.1 "Trouble In Paradise"
Released on December 8, 2012.
** Improvements
- All deprecated options since version 4.0 have been removed.
- A new primary seam-line generator, based on a Graph-Cut algorithm,
has been implemented during GSoC 2011 by Mikołaj Leszczyński. The
old Nearest-Feature Transform remains the default. The new
algorithm is activated with "--primary-seam-generator=graph-cut".
- The difference image on which Enblend's seam-line optimization
relies for color images uses a perceptual model by default (option
"--image-difference"), yielding a true CIE76 "Delta E" with the
(default) luminance and chrominance weights of 1.
- Parallelize CIECAM02 color conversion (option "--ciecam"). Speedups
of up to 40% have been reported on the amd64 architecture. The
computationally expensive color-space conversion still slows down
blending and fusing by some 25%.
- Enblend and Enfuse integrate seamlessly in color-managed workflow.
Input images with color profiles automatically enable CIECAM
blending and the output image is assigned the input images' color
profile.
- Enblend and Enfuse exploit a new feature in LittleCMS Version 2.x
called "Unbounded CMM". Thereby, the hue and saturation of extreme
shadows and highlights can be preserved much longer before pure
black or white are reached. See:
http://www.littlecms.com/CIC18_UnboundedCMM.pdf
- Assign different profiles to profile-free input images with option
"--fallback-profile" instead of being tied to sRGB.
- A new gray-scale projector called "anti-value" helps when fusing
with the intent of minimizing the noise in the output image. When
employing a lower exposure cutoff this even is the default
projector.
- Both Enblend and Enfuse stop right after saving all generated masks
to files, if option "--save-masks" is given, but option "--output"
is not. This allows to splice tools that manipulate the masks and
feed the masks back into Enblend and Enfuse with option
"--load-masks". When combining option "--output" and "--save-masks"
Enblend and Enfuse write all masks and the final output image as
before.
- Both Enblend and Enfuse can write their output JPEG files with
arithmetic JPEG compression and TIFF files with JPEG compression, if
the underlying JPEG and TIFF libraries support these compression
schemes.
** Bug Fixes
- Enblend and Enfuse consistently warn if they are passed input images
that alternate between with color profile and without.
- Grant SourceForge feature request ID 2991909 (OSX only). Previous
to that, Enblend, when launched from a terminal window with
GPU-acceleration (option "--gpu"), started a second, non-responsive
terminal window and when this new Enblend window tried to "steal"
the focus, it displayed the OSX rotating "beach ball". With the fix
that all goes away. Also the responsiveness of the screen improves,
which means that the non-functional window also hogged GUI resources.
** New Commandline Options
- Option "--primary-seam-generator" controls the primary seam-line
generation algorithm.
- Option "--image-difference" selects the difference image calculation
algorithm and optionally assigns weights to the luminance and the
chrominance part of the difference image.
- Short option "-c" now has a sibling called "--ciecam". The long
option also has a negated form: "--no-ciecam"; the short one has
not.
- The option "--fallback-profile=PROFILE" allows users to pass their
own profiles for image sets that come without ICC color profiles.
Before the hard-coded profile was sRGB.
- The option "--exposure-cutoff" facilitates tailoring the exposure
weight curve specifically to exclude underexposed (and probably
noisy) or overexposed pixels from fusion.
- In Enblend option "--load-masks" disables all mask computations and
loads the blend masks directly from the specified files. This is
the dual of Enblend's long-known option "--save-masks".
- The option "--layer-selector" overrides the standard default
layer selector.
- In addition to the usual integral values option "--levels" takes the
keyword "auto", which restores the default. (The default is to
automatically choose the maximum number of pyramid levels for each
separate overlapping region.)
** Deprecated Features
- The option "--smooth-difference" will be eliminated in the next
version of Enblend; it has not helped and moreover only put lipstick
on the seamline-optimizer's objective function.
- The user-(re)sizable image-cache will not be available in later
versions of Enblend and Enfuse. We are not sure yet whether it will
be replaced by an mmap-based solution or just tossed out.
** Developer Stuff
- XHTML validation does not rely on network access anymore. This
means all necessary DTDs must be available locally and all catalogs
must be set up correctly to build the XHTML documentation.
- The option "--parameter" allows developers to pass arbitrary
key-value pairs to Enblend and Enfuse. The keys must match the
regular expression [A-Za-z][A-Za-z0-9_-]*; the values can be almost
any string. No further checking or validation is performed.
For more explanations on how to use parameters in the source code,
see the block-comment within "namespace parameter" in file
"common.h".
** Package Maintainer Stuff
- Enblend and Enfuse now use LittleCMS version 2.x.
- Enblend and Enfuse no longer rely on their own versions of the Vigra
imaging library. Vigra version 1.8 or later is now required to
build.
- Enblend no longer relies on libXMI. (Enfuse never needed this
library.)
* Version 4.0
Released on December 13, 2009.
** Improvements
- Huge new documentation in Info, PS, PDF, and XHTML formats. Plain
HTML format is not supported because it cannot portably render
MathML.
- New manual pages.
- Automatic bit depth conversion between input and output images if
necessary.
For example: given 16-bit TIFF input images and a JPEG output image
Enblend and Enfuse automatically convert from 16-bit to 8-bit before
writing the JPEG file.
- Compilation without image cache works again
("--disable-image-cache"). Speedups of up to 30% have been reported
on the amd64 architecture.
- Enblend and Enfuse accept response files in addition to literal
image files. Response files contain lists of image filenames or
names of other response files.
- Lots of new warnings if a command-line option has no effect, like,
for example, combining a mask optimization option with
"--no-optimize".
- Enblend shows the initial, unoptimized seam line in addition to the
optimized one in all seam-line visualization images (option
"--visualize").
- Enblend and Enfuse read multi-layer TIFF files ("multi directory" in
TIFF jargon). It is even possible to mix multi-layer with single
layer images. This partially fixes SourceForge bug ID 1170329.
- Save masks with (lossless) "Deflate" compression.
- An improved wrap-around option, formerly only "-w", now "-wMODE" and
"--wrap=MODE", lets the user determine whether to create a 360
degrees horizontal or vertical panorama. It is possible to wrap
around vertically and horizontally at the same time.
- Without any output filename given, i.e. no "-o" option, the output
filename now defaults to "a.tif", making "-o" truly an option.
- The user has better control of the number of pyramid levels.
Previously only the maximum number in any blend could be reduced
by the "-l" option. Now, the maximum number can be reduced by a
fixed amount no matter what the actual maximum number of permissible
pyramid levels is.
- Use OpenMP to make better use of multi-processor machines. As the
image cache is not reentrant, OpenMP can only be activated when the
image cache is disabled.
- Enblend and Enfuse issue messages following the GNU standard.
** Bug Fixes
- Fix the calculation of the number of blending levels. Previously,
it was off by one in most cases.
- A longstanding bug (SourceForge ID 2160427) in the primary seam line
generation that caused a sub-optimal seam line was fixed. As a side
effect the new seam line algorithm is at least 20% faster than the
old one.
- Another longstanding bug (SourceForge ID 1891785) that caused wrong
output for images with holes was also fixed.
- Fix a well hidden but serious bug in the seam line optimization.
Both, the GPU and non-GPU versions of the optimizer were affected.
- Make Enblend and Enfuse aware of the input images' resolution. Use
the input resolution in the output file.
With this change masks get the correct offsets. This fixes the bug
that masks do not end up at the correct position with respect to the
input files.
- Fix MacOS problem of parsing floating-point parameters at the
command line.
** New Commandline Options
- All Enfuse options have been converted to Enblend's option naming
scheme. This is, camel-casing has been replaced by lowercase plus
dashes. The new option name are more systematically constructed,
too. The old options will completely disappear in the next release,
4.1.
Old Option New Option
==================== ========================
--ContrastWindowSize --contrast-window-size
--EdgeScale --contrast-edge-scale
--EntropyCutoff --entropy-cutoff
--EntropyWindowSize --entropy-window-size
--GrayProjector --gray-projector
--HardMask --hard-mask
--MinCurvature --contrast-min-curvature
--SaveMasks --save-masks
--SoftMask --soft-mask
--wContrast --contrast-weight
--wEntropy --entropy-weight
--wExposure --exposure-weight
--wExposureMu --exposure-mu
--wExposureSigma --exposure-sigma
--wSaturation --saturation-weight
- New option to control the output bit depth ("--depth").
- Mask filenames can be specified with templates.
* Enblend ("--save-masks", "--load-masks", "--visualize"): Save and
load all masks of a project in a single run. Same for seam line
visualization files.
* Enfuse ("--save-masks"): Save soft and hard masks according to
path given in template.
- New option to control the mask vectorization, which is done before
any mask optimizations ("--mask-vectorize").
- New option to control the Simulated Annealing optimizer called
strategy 1 ("--anneal").
- New option to control the relative weights of the distance to the
initial seam line and the total mismatch ("--optimizer-weights").
- New option to control the Dijkstra path minimizer called strategy 2
("--dijkstra").
- Option "--coarse-mask" takes an optional (linear) reduction factor.
The default remains at 8.
- New option "--smooth-difference" to blur difference images prior to
seam-line optimization.
- Long option "--levels" to control the number of levels; duplicates
functionality of short option "-l".
** Removed Commandline Options
- After having been deprecated for a long time, Enfuse's option
"--debug", which saved the masks used in fusing, was removed.
Option "--save-masks" now provides a superset of the previous
functionality.
- After having been deprecated for a long time, Enblend's and Enfuse's
option "-z", which compressed TIFF-output files with the LZW
algorithm, was removed. Option "--compression" now provides a
superset of the previous functionality and "--compression=LZW" is a
substitute for "-z".
** Developer Stuff
- VPATH builds work. This makes it easy to compile for differently
configured versions, like for example with and without image cache.
- The target "make distcheck" works again.
- Introduce additional Make variables to control the Make process
after the project was configured. See "README" file.
- Debugging of dynamic memory allocation has been eased by the
integration of the DMalloc package; "configure --with-dmalloc"
- Use only m4 files that clearly have a GPL license. This fixes
SourceForge feature request ID 2152850.
- Remove all files that are meant to be generated auto Autoconf. This
fixes SourceForge bug ID 2036031.
- Install m4 file to detect the M$ compiler. This fixes SourceForge
bug ID 2036034.
- Make "configure" report that extra features will get compiled and if
they cannot tell the developer why not.
- Pull version of program and documentation into file "VERSION". Both
program and documentation automatically use this information on
every build.
- The MSVC project files were removed. They can be regenerated by the
(still experimental) CMake environment.
- README hosts build instructions for GNU Autoconf/Automake based
systems, for MacOSX, and for Win.
** What is known not to work?
- Documentation generation is not as reliable and smooth as it ought
to be. The culprit is Automake; see PR/486.
* Version 3.2
Released on September 8, 2008.
** Improvements
- Official release of Enfuse for automatic multi-level exposure
blending and focus stack creation.
- Support for PNG and OpenEXR input files.
- Support for writing JPEG output files (8 bit only).
** Bug Fixes
- Bugfix to correctly interpret the alpha channel of float (HDR)
images.
** New Commandline Options
- New --compression option for selecting the compression type of the
output image.
** Developer Stuff
- This release includes a Windows binary that does not require a
SSE2-capable processor.
- There will not be a 3.1 release in order to avoid confusion with
numerous cvs builds that self-identify as version 3.1.
* Version 3.0
Released on January 27, 2007.
** Improvements
- New, faster image processing algorithms for computing Gaussian and
Laplacian pyramids.
** New Commandline Options
- Optional optimization of seam line placement to try to avoid
mismatches and parallax errors in the image overlap region.
- Masks can now be saved and loaded from files. This makes it
possible to manually edit the location of the seam before
multiresolution spline blending is applied.
- Checkpointing of partial results is now optional. This improves
speed.
- Optional blending using the CIECAM02 color appearance model. Your
TIFF files should have embedded ICC profiles in order to use this
option. This replaces the CIE L*a*b* color blending in Enblend 2.X
that never worked properly.
- Optional use of the graphics processor to speed up certain
computations. This feature is experimental and may not work on all
systems.
** Developer Stuff
- Like other X.0 releases, please consider this a beta.
* Version 2.5
Released on December 11, 2005.
** Bug Fixes
- Fixed a bug where Enblend would crash when the -w parameter was
used.
- Fixed a bug where Enblend would sometimes say "mask transition line
bounding box undefined."
* Version 2.4
Released on December 3, 2005.
** Improvements
- Enblend will now create output files with embedded ICC profiles.
The first ICC profile found amongst the input images will be copied
to the output image. Enblend does not use ICC profiles to do color
calculations.
- Incorporated a patch from Fulvio Senore to make the mask generation
faster. Fulvio wrote a more efficient data structure for use in the
nearest feature transform.
** New Commandline Options
- Added support for working with cropped and shifted input images.
These types of files are created by Nona's "Multiple TIFF" stitching
option. A cropped and shifted TIFF saves space and time because it
is just large enough for a single input image, instead of being the
size of the entire output panorama with lots of empty space all
around. Nona embeds an (x, y) offset coordinate in the file so that
Enblend can tell where this file belongs in the final panorama.
By default, if you give Enblend cropped and shifted TIFFs, the
output will also be a cropped and shifted TIFF. Sometimes you may
want to include the extra blank space anyway, for example if you are
creating a 360-degree panorama and the image size must be exactly a
2:1 ratio. In this case, use the new -f parameter to manually set
the size of the output image:
-f WIDTHxHEIGHT
Thanks to Pablo d'Angelo for providing this patch.
* Version 2.3
Released on April 17, 2005.
** Bug Fixes
- Fixed a bug in temporary file handling in the Windows version of
enblend. This should solve the "unable to open temporary file"
error.
- Reduced the maximum number of levels you can specify with the -l
parameter from 30 to 29. While both of these are impractically
large, at least 29 does not lead to arithmetic overflow and a
subsequent crash.
- Fixed a bug in temporary file handling in the Windows version of
enblend. This should solve the "unable to open temporary file"
error.
* Version 2.2
Released on February 5, 2004.
** Improvements
- The Windows binary of Enblend is now build with a version of libtiff
that includes support for all of the TIFF compression standards,
such as Deflate, LZW, JPEG, and Packbits. Nona/Hugin produces TIFF
files with the Deflate option.
- Replaced some system calls that are specific to Windows 2000 and XP
with more generic functions. This should get the Windows binary of
Enblend to run on pre-Win2K machines.
** Bug Fixes
- This release fixes issues with the Windows version of Enblend. If
you are using Enblend on UN*X you do not need to upgrade.
* Version 2.1
Released on November 15, 2004.
** Improvements
- Compression is no longer the default option in the Windows
executable.
** Bug Fixes
- Turned off TIFF library warning messages that required user
interaction on Windows.
- Fixed a bug in Vigra that caused primary color spots to appear in
overexposed areas of 16-bit images.
- Fixed a problem with Enblend crashing on large panoramas. Modified
tiff import to use a scanline-based interface instead of a strip-
based interface. PTStitcher generates TIFFs with the rows/strip tag
set to an unreasonable value.
** Developer Stuff
- Ported the source to compile natively on Win32 using MSVC.
* Version 2.0
Released on October 17, 2004.
** Improvements
- Support for signed and unsigned 16-bit, 32-bit, single- and double-
precision floating point pixel types.
- Sophisticated memory/disk balancing. You can tell Enblend how much
memory it is allowed to use, and it will swap to disk after that.
- Support for huge panoramas. I have tested that Enblend can blend a
1.2 gigapixel, 16-bit per channel color image. You should be able
to go right up to the 4 gigabyte limit of the TIFF format.
** Bug Fixes
- Fixed the banding artifacts that often appeared in skies.
** New Commandline Options
- Optional blending in CIE L*a*b* color space.
- Option to use LZW compression for the output image, if your libtiff
supports it. This option is selected by default in the prebuilt
Windows executable.
- The -s option for sequential blending is now the default. Enblend
cannot accurately determine how many blending levels to use when
there are multiple disjoint overlap regions in one blending
iteration. It is better to manually tell Enblend the order in which
the images should be blended. If you still want Enblend to assemble
non-overlapping images first, use the -a flag.
- The -l option now tells Enblend the exact number of blending levels
that you want to use. The geometry of your images may force Enblend
to use a smaller number of levels. In this case a message will be
printed.
- Gimp (ver. < 2) and Cinepaint exhibit unusual behaviors when loading
images with unassociated alpha channels. Use the -g flag to work
around this. With this flag Enblend will create the output image
with the associated alpha tag set, even though the image is really
unassociated alpha.
** Developer Stuff
- Ported Enblend to the VIGRA Computer Vision Library.
* Version 1.3
Released May 18, 2004.
** Bug Fixes
- Added a configure test for the endian-ness of the machine. I
changed the way enblend reads and writes 8-bit color fields in
32-bit pixel words to reflect the endian-ness of the machine. This
should get enblend working on powerpc machines.
This release adds no new features.
* Version 1.2
Released April 26, 2004.
** Bug Fixes
- Fixed a bug where 8-bit TIFFs would not be identified properly on
big-endian machines.
- A second (better) fix for the boundary condition problem that lead
to horizontal or vertical seams to appear in the output.
* Version 1.1
Released April 3, 2004.
** Bug Fixes
- Fixed a bug in reading from temporary files that appeared on
machines with different versions of GCC.
- Fixed a boundary condition problem that lead to horizontal or
vertical seams to appear in the output.
- Fixed a bug where segments of an input image would appear on the far
side of a blending zone. This would cause strange wedge-shaped
artifacts in some output images.
* Version 1.0
Released March 27, 2004.
** New Commandline Options
- Added options to limit the number of blending levels used. This can
reduce memory usage and improve speed, at a loss of quality.
- Added an option to blend around the -180/+180 degree boundary.
There is still an open issue as to how to handle the zenith and
nadir.
- Added an option to blend the input images sequentially in the order
given on the command line. Sometimes the assembler can make poor
choices and seams will still be visible in the output.
** Developer Stuff
- Replaced erosive thinning algorithm for mask creation with a much
faster nearest feature transform.
- Reduced memory requirements. Enblend now swaps most of its data to
disk as temporary files. Enblend also calculates the region of
pixels that are involved in each blending step and only does math on
those pixels, improving speed and memory usage.
* Version 0.9
Released on March 4, 2004.
Local Variables:
mode: outline
End: