Your IP : 216.73.216.86
PK k:�\?ٰ7 7 ANNOUNCEnu �[���
JBIG-KIT lossless image compression library
-------------------------------------------
by Markus Kuhn
The latest release of JBIG-KIT can be downloaded from
http://www.cl.cam.ac.uk/~mgk25/jbigkit/
JBIG-KIT implements a highly effective data compression algorithm for
bi-level high-resolution images such as fax pages or scanned
documents.
JBIG-KIT provides two variants of a portable library of compression
and decompression functions with a documented interface. You can very
easily include into your image or document processing software. In
addition, JBIG-KIT provides ready-to-use compression and decompression
programs with a simple command line interface (similar to the
converters found in Jef Poskanzer's PBM graphics file conversion
package).
JBIG-KIT implements the specification
International Standard ISO/IEC 11544:1993 and ITU-T Recommendation
T.82(1993), "Information technology - Coded representation of picture
and audio information - progressive bi-level image compression",
<http://www.itu.int/rec/T-REC-T.82>,
which is commonly referred to as the "JBIG1 standard". JBIG (Joint
Bi-level Image experts Group) is the committee which developed this
international standard for the lossless compression of images using
arithmetic coding. Like the well-known compression algorithms JPEG and
MPEG, JBIG has also been developed and published by the International
Organization for Standardization (ISO) and the International
Telecommunication Union (ITU). See also
http://www.jpeg.org/jbig/
http://www.iso.ch/
http://www.itu.int/
The JBIG compression algorithm offers the following features:
- Close to state-of-the-art lossless compression ratio for high
resolution bi-level images.
- About 1.1 to 1.5 times better compression ratio on typical
scanned documents compared to G4 fax compression (ITU-T T.6),
which has been the best compression algorithm for scanned
documents available prior to JBIG.
- Up to 30 times better compression of scanned images with dithered
images compared to G4 fax compression.
- About 2 times better compression on typical 300 dpi documents
compared to 'gzip -9' on raw bitmaps.
- About 3-4 times better compression than GIF on typical 300 dpi
documents.
- Even much better competitive compression results on computer
generated images which are free of scanning distortions.
- JBIG supports hierarchical "progressive" encoding, that means it is
possible to encode a low resolution image first, followed by
resolution enhancement data. This allows, for instance, a document
browser to display already a good 75 dpi low resolution version of
an image, while the data necessary to reconstruct the full 300 dpi
version for laser printer reproduction is still arriving (say
over a slow network link or mass storage medium).
- The various resolution layers of a JBIG image in progressive
encoding mode together require not much more space than a
normal non-progressive mode encoded image (which JBIG also
supports).
- The progressive encoding mode utilizes a quite sophisticated
resolution reduction algorithm which offers high quality low
resolution versions that preserve the shape of characters as well
as the integrity of thin lines and dithered images.
- JBIG supports multiple bit planes and can this way also be used
for grayscale and color images, although the main field of
application is compression of bi-level images, i.e. images with
only two different pixel values. For grayscale images with up to
6 bit per pixel, JBIG performs superior to JPEG's lossless
mode.
JBIG-KIT can be used as free software under the GNU General Public
License. Other license arrangements more suitable for commercial
applications are available as well, please contact the author for
details. JBIG-KIT provides two portable libraries implemented in
ANSI/ISO C for encoding and decoding JBIG data streams, along with
documentation. The first library, jbig.c, implements nearly all of the
options that the JBIG standard provides, but keeps the entire
uncompressed image in memory. The second library, jbig85.c, implements
only the ITU-R T.85 subset of the standard that black/white fax
machines use (single bit per pixel, no "progressive" encoding), and
keeps only three lines of the uncompressed image in memory, making it
particularly attractive for low-memory embedded applications.
The libraries are not intended for 8-bit or 16-bit machine
architectures (e.g., old MS-DOS C compilers). For maximum performance,
a 32-bit processor is required (64-bit systems work too, of course).
On architectures with 16-bit pointer arithmetic, the full-featured
jbig.c library can process only very small images.
Special features of the full-featured jbig.c variant:
- Fully reentrant multithread-capable design (no global or static
variables, isolated malloc()/free() calls, etc.)
- Capable of handling incomplete and growing JBIG data streams in
order to allow earliest display of low resolution versions
- Capable of handling several incoming data streams simultaneously
in one single process and thread
- Especially designed with applications in mind that want to display
incoming data as early as possible (e.g., similar to the way in
which Netscape Navigator handles incoming GIF images)
- Implements all JBIG features and options including progressive and
sequential encoding, multiple bit planes, user specified
resolution reduction and deterministic prediction tables, adaptive
template changes for optimal performance on half-tone images,
deterministic prediction, typical prediction in lowest and
differential layers, various stripe orderings, etc; only the SEQ
and HITOLO options are currently not supported by the decoder
(they are normally never required, but could be added later in
case of user requirements)
- Suitable for fax applications, satisfies ITU-T T.85 profile
- Efficient code, optimized utilization of 32-bit processor
registers
- Very easy to use documented C library interface
- Included Gray code conversion routines for efficient encoding
of grayscale images
- Ready-to-use pbmtojbg and jbgtopbm converters.
Special features of the light-weight jbig85.c variant:
- Suitable for low-memory embedded applications
- Implements only the JBIG1 subset defined in the ITU-T T.85
profile (single bit plane, no differential layers)
- Requires only three pixel rows of the uncompressed image to be
kept in memory
- Handles all NEWLEN modes of operation required by ITU-T T.85 with
just a single pass over the data, automatically performing the
necessary lookahead after the last stripe
- Codec buffers only a few bytes of arithmetic-codec data and outputs
resulting bytes or lines as soon as they are available.
I will try to provide free support and maintenance for this software
for the foreseeable future, depending on my available time.
Happy compressing ...
Markus Kuhn
--
Markus Kuhn, Computer Laboratory, University of Cambridge
http://www.cl.cam.ac.uk/~mgk25/ || CB3 0FD, Great Britain
PK k:�\a3���# �# CHANGESnu �[���
JBIG-KIT revision history
-------------------------
Changes in version 2.1 (2014-04-08)
This is a security-critical bug-fix release that remains API and ABI
backwards compatible to version 2.0. Users who process BIE data from
untrusted sources should upgrade.
- fixed a buffer-overflow vulnerability in the jbig.c decoder,
reported by Florian Weimer (Red Hat): CVE-2013-6369
- fixed ability of corrupted input data to force jbig85.c decoder
into an end-less loop
- fixed a bug in the processing of private deterministic-prediction
tables (DPPRIV=1) in jbig.c decoder
- fixed integer-type mismatches in printf arguments on 64-bit systems
- fuzz-testing script added
Changes in version 2.0 (2008-08-30)
Main new features of release 2.0:
- This release adds a separate lightweight "T.85" version of the
encoder and decoder library (jbig85.c, jbig85.h). This new T.85
library is optimized for embedded low-memory applications, such as
printers and fax machines. It implements only the subset of the
JBIG standard that is required by the ITU-T T.85 fax profile,
namely only a single bit plane (P = 1) and no differential layers
(DL = D = HITOLO = SEQ = ILEAVE = SMID = TPDON = DPON = DPON =
DPLAST = 0) and some other restrictions (MY = 0, MX < 128).
The T.85 library requires only three pixel rows of the
uncompressed image to reside in memory at any time. This 3-line
buffer has to be allocated by the calling application, therefore
the T.85 library will not call any heap management functions such
as malloc() or free() itself.
- There are also two new example applications pbmtojbg85 and
jbgtopbm85 included that provide command-line access to the new
T.85 library. (Note to distribution maintainers: there is probably
little point in adding pbmtojbg85 and jbgtopbm85 to precompiled
JBIG-KIT packages for workstation operating systems, as these two
programs do not really provide the end user much additional
functionality beyond what pbmtojbg and jbgtopbm already do, which
use the full library instead.)
Changes to the original full jbig.c library from version 1.6:
- the arithmetic codec routines are now located in a separate source
file (they are now used by both the full library jbig.c and the new
T.85 library jbig85.c, and might also be of use to other applications)
- arithmetic decoder behaviour slightly changed to enable the
lookahead needed to handle T.85-style NEWLEN after the final
stripe
- added encoder support for adding comment marker segment
- added encoder option for using SDRST marker instead of SDNORM
(only useful to make encoder able to generate a wider range
of test data)
- added pbmtojbg options -C and -r, which provide command-line
access to the previous two extensions
- slightly improved diagnostic output of jbgtopbm option -d
- non-English translations of error messages are now provided
in a separate PO file (e.g., for use with GNU gettext),
jbg_strerror() no longer has a language attribute and the
JBG_EN, JBG_DE_8859_1, JBG_DE_UTF_8 constants no longer exist
- jbg_dec_in() return values changed; the least significant bits of
the error codes now provide more information about what caused the
error, beyond what the more general error text messages returned by
jbg_strerror() reveal
- pbmtojbg has a new option -f as a shortcut for all the options
needed to make the output comply to the basic ITU-T T.85 fax
profile
- jbg_dec_getwidth(), jbg_dec_getheight(), jbg_dec_getsize(), and
jbg_dec_getsize_merged() return now an unsigned long value
(was: long), and return 0 (was: -1) if no image is available yet
- jbgtopbm now outputs image dimensions in header padded to 10
digits fixed width, for consistency with jbgtopbm85
to allow for later in-place update of height due to NEWLEN
- macro JBG_LICENCE can be changed from the default value "GPL" to a
licence agreement reference code to if the library is used under a
commercial licence, to clarify under which exact licence the
library is used in a particular application
Changes in version 1.6 (2004-06-11)
- various small changes to reduce the risk of 32-bit unsigned
integer overflows when dealing with extremely large images
- robuster treatment of L0 = 0xffffffff
- minor API modification in jbg_enc_options(): parameter l0 changed
from type long to unsigned long; previous value now remains
unchanged when l0 == 0 (was: l0 < 0)
- lots of type casts added such that the C source code is now
also compilable as C++
Changes in version 1.5 (2003-06-11)
- fixed two minor memory leaks (special thanks to Davide Pizzolato
<ing.davide.pizzolato@libero.it> for locating one of these)
- jbgtopbm does not attempt any more to parse multiple concatenated
BIEs (options -m must be used now to enable this feature explicitly),
in order to handle BIEs with data after the last expected SDE gracefully
- various extensions to improve support of JBIG fax applications
(ITU-T T.85 profile):
o support for maximum adaptive template pixel offset increased
to MX=127 in both encoder and decoder
o encoder now has a hook for producing BIEs with a NEWLEN marker
segment and VLENGTH=1, in order to assist in testing decoders
for T.85 conformance (see also new pbmtojbg option -Y)
o a new function jbg_newlen() can be used to scan an
entire in-memory BIE for NEWLEN marker segments and update the
YD value in the BIE header, which can be applied to BIE data
before passing it to the decoder for T.85 compliance
(Background: the incremental design of the JBIG-KIT decoder
makes it infeasible to look ahead for NEWLEN marker segments
that occur after the SDE with the last image line, therefore
this second pass is necessary to handle the old-style fax
applications permitted by T.85 in which the low-RAM encoder
doesn't know the height of the image at the start of
transmission)
Changes in version 1.4 (2002-04-09)
- typo fixed in stripe number calculation, which caused encoder and
decoder to fail on certain image sizes (special thanks to Koen Denecker
<denecker@elis.rug.ac.be> for locating this one)
Changes in version 1.3 (2002-03-23):
- bugs fixed in jbg_split_planes(), jbg_dec_merge_planes() that
caused a failure for more than eight planes per pixel
- example converters now can handle PGM files with up to 32 bits
per pixel
Changes in version 1.2 (2000-04-08):
- bug in the decoder fixed, which caused the rest of the input file
to be skipped whenever a comment marker was encountered (special
thanks to Ben Rudiak-Gould <benrg@math.berkeley.edu> for
reporting this one)
Changes in version 1.1 (1999-11-16):
- serious bug in the encoder fixed, which for a very small
percentage of images has caused an unterminated linked list to be
created internally that could have been responsible for
segmentation violations or non-terminating encoders
(special thanks to Hisashi Saiga <saiga@itl.tnr.sharp.co.jp> for
tracking that one down)
- minor bug in the "jbgtopbm -d" diagnostic output fixed
Changes in version 1.0 (1998-04-11):
- two bugs fixed that caused the encoder and decoder to fail
under certain modes of operation with several bit planes
- added new functions jbg_split_planes(), jbg_dec_merge_planes(),
and jbg_dec_getsize_merged() for easy handling of grayscale
images
- added support for compressing grayscale PGM files to pbmtojbg
and jbgtopbm
- more changes to avoid paranoid compiler warnings
Changes in version 0.9 (1996-01-09):
- encoder won't break any more on input bitmap data with incorrect
zero padding
- pbmtojbg displays a warning if input file has incorrect zero
padding
- various minor improvements suggested by Stefan Willer
<Stefan.Willer@unnet.wupper.DE>
- many minor changes in order to avoid warnings from paranoid
compilers
Changes in version 0.8 (1995-09-20):
- namespace cleared up, all names externally visible from the library
start now with jbg_ or JBG_
- minor non-critical bug fixed which caused library to fail compatibility
test and showed up especially on DEC Alpha systems
- jbg_dec_gethight() is now called jbg_dec_getheight()
- filenames conform now to MS-DOS limits
- Bug in pbmtojbg fixed (handling of ASCII PBM files)
Changes in version 0.7 (1995-06-10):
- more problems on 16-bit int systems and on Macintosh systems fixed
(special thanks to Jean-Pierre Gachen <jpg11@calvanet.calvacom.fr>)
- global Makefile
Changes in version 0.6 (1995-06-08):
- memory leak fixed
- should now also work on systems where int is only 16-bit large
- changes of the JBIG "Technical Corrigendum 1" included (special
thanks to Dr. Sebestyen from Siemens AG for sending me a copy
of the draft)
First release: version 0.5 (1995-05-28)
PK k:�\���] ] TODOnu �[��� Features that may appear in some future release regarding jbig.c:
- function to prepare incompletely decoded images for display by Xlib
- use a more efficient resolution-reduction method suggested by
Dr. Klaus Bartz <bartzkau@kou3.ina.de>
- investigate whether there is a standard way of embedding JBIG
into TIFF and implement it (see also RFC 2301)
- test for and eliminate warning messages from compilers
other than GCC
- multichannel (e.g., RGB) support, PPM support
- add dynamic library target to Makefile (patch by David Woodhouse),
for the benefit of Linux packagers
The NEWLEN handling in jbig.c remains less than ideal. There is no
easy fix as NEWLEN is inherently incompatible with many of the other
options that jbig.c supports. The whole point of the VLENGTH bit may
have been to allow recipients to choose between a jbig.c (VLENGTH=0)
and jbig85.c (VLENGTH=1) style of decoder implementation. The cleanest
option would really be to remove NEWLEN support entirely from jbig.c
and to urge users who need it to use jbig85.c instead. Short of that,
the new NEWLEN look-ahead architecture from jbig85.c could be
backported to jbig.c, but this would only only slightly increase the
number of cases where the jbig.c decoder could deal with a NEWLEN
marker without a prior invocation of the jbg_newlen() pre-processing
pass.
The following past feature requests for jbig.c probably have become
less important since the release of jbig85.c:
- compile-time option to remove progressive encoding and
resolution-reduction support (in the interest of reducing code size)
- investigate how to best deal with broken BIEs that contain
multiple NEWLEN marker segments, which are not permitted by ITU-T
T.85, but which have been observed coming from some fax machines;
possible options for jbg_newlen(): use first, last, or minimum
value
- version of jbg_newlen() for dealing with BIEs that are split
into several blocks
- call-back function for progress meter
Features that may appear in some future release regarding jbig85.c:
- man pages for pbmtojbg85 and jbgtopbm85
PK k:�\?ٰ7 7 ANNOUNCEnu �[��� PK k:�\a3���# �# o CHANGESnu �[��� PK k:�\���] ] a@ TODOnu �[��� PK � �H