Low bit rate compression performance (rates below 0.25 bpp for highly-detailed gray-level images)
	Lossless and lossy compression in a single code stream
	Seamless quality and resolution scalability, without having to download the entire file. The major benefit is the conservation of bandwidth
	Large images: JPEG is restricted to 64kx64k images (without tiling). JPEG2000 will handle image sizes up to (2^32 - 1)
	Single decompression architecture
	Error resilience for transmission in noisy environments, such as wireless and the Internet
	Computer generated imagery
	Compound documents
	Region of Interest coding
	Improved compression techniques to accommodate richer content and higher resolutions
	Metadata mechanisms for incorporating additional non-image data as part of the file

JPEG2000 will be able to handle up to 256 channels of information, as compared to JPEG, which is limited to only RGB data. Thus, JPEG2000 will be capable of describing complete alternate color models, such as CMYK, and full ICC (International Color Consortium).

Compression Efficiency

Early results show a 20% compression efficiency improvement over JPEG, and a 40% improvement over Flashpix.

Important factors taken into account for achieving high compression efficiency:

	Embedded lossy to lossless
	Multiple component images
	Static and dynamic Region-of-interest
	Error resilience
	Spatial and quality scalability
	Rate-control

JPEG2000 has two coding modes:

	DCT-based coding mode: Currently baseline JPEG
	Wavelet-based coding mode: Includes non-reversible and reversible transforms

For an complete definition of the existing JPEG2000 compression system, please refer to the Verification Model (see "References" below). The discussion below is adapted from the JPEG2000 VM4.0.

Overview

The coder is essentially a bit-plane coder, using the same Layered Zero Coding (LZC) techniques which have been employed in a number of embedded Wavelet coders and were originally proposed by Taubman and Zakhor (IEEE Tx IP, Sep '94). (See References below.) In fact, many of the ideas presented in the VM4.0, including the use of separate code blocks and post-compression rate-distortion optimization are taken directly from that work and Dr. Taubman's own doctoral dissertation (UC Berkeley, 1994). The key additions are:

	The use of fractional bit-planes, in which the quantization symbols for any given quantization layer (or bit-plane) are coded in a succession of separate passes, rather than just one pass
	A simple embedded quad-tree algorithm is used to identify whether or not each of a collection of "sub-blocks" contains any non-zero (significant) samples at each quantization layer, so that the encoding and decoding algorithms need only visit those samples which lie within sub-blocks which are known to have significant samples.

EBCOT: The Basic Idea

In VM4, the coding algorithm used is known as "Embedded Block Coding with Optimized Truncation" (EBCOT). The coding subsystem in JPEG2000 is responsible for both the low level entropy coding operations associated with the representation of subband sample values, and organizing and packing the resulting code words into the bit stream.
The basic idea in EBCOT is to divide each subband into blocks of samples which are coded independently. For each block, a separate bitstream is generated without using any information from the other blocks. The bit stream has the property that it can be truncated to a variety of discrete lengths.

Once the entire image has been compressed, a postprocessing operation passes over all the compressed blocks and determines the extent to which each block's embedded bit stream should be truncated in order to achieve a particular target bit rate, distortion bound or other quality metric. More generally, the final bit stream is composed from a collection of so-called "layers", where each layer has an interpretation in terms of overall image quality.

The first, lowest quality layer, is formed from the optimally truncated block bit streams in the manner described above. Each subsequent layer is formed by optimally truncating the block bit streams to achieve successively higher target bit rates, distortion bounds or other quality metrics, as appropriate, and including the additional code words required to augment the information represented in previous layers to the new truncation points.

An important aspect of the EBCOT algorithm is the manner by which it forms a final bit stream from the independent embedded bit streams generated for every block. The bit stream formation problem is very much simplified when the coder operates on entire subbands at a time, since the additional spatial organization imposed by independent blocks does not exist.

The fact that blocks are encoded independently enables the "random access" feature. Suppose, however, that the bit stream must also possess the "SNR progressive" feature. These two features appear to work against each other since the random access feature requires that individual blocks be separately decodable, while the SNR progressive feature requires that the embedded bit streams for these blocks be distributed throughout the bit stream so that more important information always precedes less important information, regardless of the spatial location associated with this information. It would seem that the amount of overhead required to identify the individual blocks within this distributed representation would be quite considerable.

As the image becomes larger, the increased overhead required to identify the exact sequence of a larger number of block segments is largely wasted because many of these blocks will have almost identical rate-distortion slopes so that the order in which they appear is largely immaterial. It makes sense, therefore, to identify the block truncation points which are very similar and include the relevant code bytes for each of these blocks in a pre-defined order. This is essentially the bit stream layering idea.

Basically, the bit stream is organized as a succession of layers, where each layer contains the additional contributions from each code block (some contributions may be empty). The block truncation points associated with each layer are optimal in the rate-distortion sense, which means that the bit stream obtained by discarding a whole number of least important layers will always be rate-distortion optimal. If the bit stream is truncated part way through a layer then it will not be strictly optimal, but the departure from optimality can be small if the number of layers is large.

As the number of layers is increased so that the number of code bytes in each layer is decreased, the rate-distortion slopes associated with all block truncation points in the layer will become increasingly similar; however, the number of code blocks which do not contribute to the layer will also increase so that the overhead associated with identifying the code blocks which do contribute to the layer will increase. In practice, we find that optimal compression performance for SNR progressive applications is achieved when the number of layers is approximately twice as large as the number of sub-bit-plane passes made by the entropy coder (that is, the bit stream contains twice as much granularity as that provided by previous verification models).

The boundaries of the sub-bit-plane passes are also the truncation points for each block's embedded bit stream. Consequently, on average each layer contains contributions from approximately half the code blocks so that the cost of identifying whether or not a block contributes to any given layer (about 2 bits per block) is much less than the cost of identifying a strict order on the block contributions. Moreover, the relative contribution of this overhead to the overall bit rate is independent of the size of the image.

The DIG2000 File Format Proposal

The goal of the DIG2000 Initiative is to create a digital file format that embodies a tightly-integrated set of essential features for storing images, and provides the needed mechanisms for images to be used effectively.

Some of the most important features are:

	Flexible metadata architecture
	Unambiguous specification of color (default sRGB)
	Resolution-independent coordinate system
	Asymmetric storage and delivery - server holds original and all metadata.
	Client can select subset to be delivered
	Protection of intellectual property - requires encryption, watermarking etc
	Improved quality and rendition - consistent colour, print capability
	Some backwards compatibility with original JPEG compression
	Object oriented functionalities (coding, information embedding, a€|)
	File format

Interesting links

For your convenience, these links will open in a new window.

	From the official JPEG web site: JPEG links JPEG2000links, including a PDF version of the first Committee Draft Public Relations
	A set of JPEG2000 tools for coding and decoding, JP2 file parsing and validation, and Static ROI setting and displaying.
	A non-technical article about JPEG2000 from WebReview.com
	Organization of the JPEG2000 committee, from the SPEAR project web page
	Watermarkingon JPEG2000
	The Digital Imaging Group(DIG) web site. The DIG2000 working group proposed a file format for use with JPEG2000

References

	JPEG2000 Verification Model 4.0, ISO/IEC JTC 1/SC 29/WG 1, Charilaos Christopoulos (Ericsson, Sweden), Editor, April 22 1999. (Note: The latest version of the VM is v5.0).
	Requirements Ad Hoc Group, "JPEG2000 requirements and profiles version 6.0," WG1 Vancouver Meeting, July 1999.
	D. Taubman, "High Performance Scalable Image Compression with EBCOT", to appear in IEEE Trans. Image Proc., Submitted March 1999; Revised August 1999. Available in PDF format
	D. Taubman and A. Zakhor, ``Multirate 3-D Subband Coding of Video,'' IEEE Transactions on Image Processing, September 1994, vol. 3, no. 5, pp. 572-588.
	D. Taubman, ``Directionality and Scalability in Image and Video Compression,'' Ph.D. Thesis, Department of Electrical Engineering and Computer Sciences, University of California at Berkeley, December 1994.
	JPEG2000 Committee Draft Version 1.0, December 1999. Available in PDF format
	Tutorial on JPEG2000, by Dr. Charilaos Christopoulos, presented at ICIP '99
	Video Technology Branch, Media Technologies Laboratory, DSP Solutions R&D Center, Texas Instruments.
	Digital Imaging Group, "DIG2000 file format proposal overview," DIG2000 Working Group, October 30, 1998.
	The Digital Imaging Group's DIG2000 Initiative, "An Overview of JPEG2000 Technology and Benefits,"
	JPEG Public Relations press releases