关于JPEG2000软件

本文探讨了JPEG2000编码方式的复杂性,并对比了Kakadu与其他开源软件如OpenJPEG和Jasper的性能。尽管JPEG2000的技术难度超过H.264,但作者推荐了几个开源项目供读者研究学习。

  现在网上JPEG2000的Kakadu 2.23源代码似乎找不到了。这个不奇怪,因为Kakadu本身早已商业化。其他的JPEG2000的源码也不少,例如OpenJPEG和Jasper,都是开源的。比较实验一般认为Kakadu的图像质量不好,但我们实验下来,Kakadu(2.23版本,开启MMX和汇编)在编解码速度远优于其他大部分JPEG2000软件。
  JPEG2000是一种复杂的图像编码方式,笔者可以毫不夸张地说,其技术上的复杂性甚至(远)超过H.264(表现在小波变换,位平面编码,质量层分割,更夸张的基于上下文的算术编码),当然追求目标,实现效果以及应用范围和前景另当别论。或者可以这样理解,如果真能理解JPEG2000,那么对于理解H.264,只要加上对运动和帧内预测的一些概念外,剩下的简直是雕虫小技。而资源(指支持工程开发方面的)的缺乏和Kakadu代码风格给这种复杂度雪上加霜。因此看JPEG2000代码需要很大的耐心(另外需要不少知识储备),笔者认为,这和看操作系统内核,游戏核心引擎一样,是一种闭关修炼的过程。可惜笔者在JPEG2000方面还是理解较浅,大有半途而废之势。
  推荐:
  Jasper主页:http://www.ece.uvic.ca/~mdadams/jasper/
  OpenJPEG主页:http://www.openjpeg.org/
  Kakadu 2.23版:http://www.snapdrive.net/files/199759/source/kdu2.23.rar

  至于JPEG,虽然和JPEG2000是一个名字,但是从手段本身来说差别太大。但它非常了不起,是DCT系编码手段的经典原型。一般可以用开源的库libjpeg:

Changes from version 2.2.2 to version 2.2.3 ------------------------------------------- * Extremely minor changes to avoid mixed use of formatted and unformatted calls to "ostream" objects. These appear to excite a bug in GCC version 3.0. The only file affected is "params.cpp" in "coresys/parameters". Changes from version 2.2.1 to version 2.2.2 ------------------------------------------- Note: none of these have any impact whatsoever on executable code or DLL's. * Renamed the "core" directory as "coresys". A trivial change and my appologies for those whom this might adversely affect. However, the use of the name "core" was causing some people difficulties, since it is identical to the name of a Unix core dump file. * Made the Linux MMX-optimized functions "extern C" instead of "extern", so as to avoid problems caused by different name mangling conventions between various versions of gcc. * Eliminated multi-line comments from assembler files so as to avoid problems created by earlier versions of the gnu assembler. Changes from version 2.2 to version 2.2.1 ----------------------------------------- * Replaced the C++ I/O routines used for image and compressed data transfers with ANSI C I/O functions. This was motivated by the fact that the new-style ANSI C++ I/O package is unbelievably slow, at least as implemented in Microsoft's Visual C++ compiler. The change has no impact whatsoever on the Kakadu core system; it affects only the implementation of a few application level objects -- the core system abstracts all such I/O considerations through interface classes which are implemented by applications. Everything now runs a little faster than it did in version 2.1 and quite a bit faster than it did in the first release of version 2.2. * Made provision for compiling under versions of GCC prior to version 3.0. To use this, you should define GCC_VERSION_LESS_THAN_3. Changes from version 2.1 to version 2.2 --------------------------------------- * Extensive support for ROI (Region of Interest) specification at encode time (see "kakadu.pdf" for more on this). * Migrated from the old-style C++ iostream package to the new standard iostream package -- minimal use of "using namespace std" and never used in common header files, so this should enhance namespace protection and portability. * Added AT&T style versions of the small amount of Pentium assembly code to enable compilation of a high speed version under GCC as well as MSVC. * Some minor bug fixes. Changes from version 2.0.2 to version 2.1 ----------------------------------------- * Extensive support for working with the JP2 file format. The "kdu_show" application demonstrates the capabilities required of a conformant JP2 reader: palette mapping; interpolation of components to a common resolution; application of registration offsets in the CRG marker segment; and colour conversion to an appropriate rendering space (sRGB here). The "kdu_region_decompressor" object provides extensive support for general purpose interactive rendering applications, performing all of the above tasks in a platform independent manner. * It is now possible to directly control rate-distortion slope thresholds used in the construction of quality layers. This capability may also be used to significantly increase compression speed, if a suitable threshold is known, since the encoder then incrementally predicts the point at which there is no point in coding further coding passes. * A number of improvements to the "kdu_show" application, including the ability to arbitrarily zoom into images. * A number of minor bug fixes, including one important bug reported by Aaron Deever of Kodak, and a bug which occasionally manifested itself in the incremental rate prediction heuristic (reported by Jim Andrew of CISRA). * Improved documentation. Changes from version 2.0.1 to version 2.02 ------------------------------------------ * A PDF document (documentation.pdf) has been prepared to guide the new user into some of the more important aspects of the Kakadu system. The first draft is included here. * A very simple compression example and a very simple decompression example have been added to assist developers in familiarizing themselves with the Kakadu system -- the existing demo apps provide perhaps too much functionality to be quickly understood. * A full BIBO (Bounded Input Bounded Output) numerical analysis of the DWT and other processing steps is used to establish the best utilization of limited precision sample data representations. The new version should not be able to fall prey to numerical overflow or underflow difficulties under any circumstances (this could just have been possible with the last version). It also provides slightly higher accuracy. * The automatic heuristic for generating quality layer rate targets has been substantially improved. * A number of minor bugs/limitations were corrected, although these had not manifested themselves in any real examples. Changes from version 2.0 to version 2.01 ---------------------------------------- * One line change in each of "kdu_expand.cpp" and "kdu_compress.cpp" to correct a minor rare bug in these demo applications. * Minor changes in "kdu_show.cpp" to correct a rare bug and improve the user interface in regard to image rotation. * Four lines added to each of "encoder.cpp" and "decoder.cpp" to fix a minor memory leak.
评论 9
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值