Introduction to JPEG 2000 image coding standard

Table des matières

1. Introduction to JPEG 2000

Since the development of JPEG at the beginning of the nineties, which has became the most well known and used standard for still images compression, a new standardisation work was led by the JPEG consortium to establish a new standard based on a wavelet transform rather than on a DCT transform. Denoted by JPEG 2000, the first parts of this standard were established in 2000. Improvements were obtained when compared to JPEG, in particular JPEG 2000 offers or introduces:

Nowadays, it is dubious that JPEG 2000 could replace JPEG in low complexity applications in the compression range where JPEG is efficient. On the other hand, it is plausible that the error resilience tools of JPEG 2000, as well as the error protection and correction tools offered by JPWL (JPEG 2000 Part 11) are good candidates for still images transmissions over error prone wireless channels.

2. Description of the JPEG 2000 image Coding Format

2.1 JPEG 2000 image coding format

The generation of the codestream in JPEG 2000 is illustrated in Figure 1.

Figure 1 - Generation of a JPEG 2000 codestream.

The two main originalities of the standard is first to rely on a wavelet decomposition (see Figure 2 for an illustration) and then to have the compression function achieved by arithmetic coding.

Figure 2 - Illustration of a Wavelet decomposition over three levels.

The reader interested in full details of the standard can refer him/herself to the complete standard [1]. We will here focus on the organisation of the standard and precise a few details of its syntax.

As illustrated by Figure 3, the codestream beginning with a Main header, containing general information for the whole codestream and various tile-parts. A tile-part is defined as a portion of the codestream with compressed image data for some, or all, of a tile. The codestream ends then by a marker called EOC for End of Codestream.

Figure 3 - General syntax of a JPEG 2000 codestream.

In practice, the different blocks depicted in Figure 3 are made of dedicated stream segments called Marker Segment, which all have a common syntax presented in Figure 4. The first 16 bits of the marker segment, which consists in the marker are the label which will allow the decoder to discriminate the role of the upcoming marker segment and allow to process it correctly. The second set of 16 bits, which indicates the length of the marker segment minus the header allows a decoder to have knowledge of the number of parameters used in the marker segment ... but also to skip if the decoder can not interpret it.

Figure 4 - JPEG 2000 marker segment format.

2.2 Marker segments in JPEG 2000

Table 1 lists most of the different marker segments in JPEG 2000 syntax, and gives for each their full name, the marker short name and its corresponding 2 bytes code, and finally indicates in which header it can be found or not. In this table, required means that the marker or marker segment shall be in this header, whereas optional means it may be used but is not mandatory.

• The POC marker segment is required if there are progression order changes.
• Either the PPM or PPT marker segment is required if the packet headers are not distributed in the bit stream. If the PPM marker segment is used then PPT marker segments shall not be used, and vice versa.

The place and possible order of the markers is illustrated in Figure 5, Figure 6 and Figure 7 respectively for the Main Header, the first Tile-part of a Tile and any non-first Tile-part of a tile.

Figure 5 - JPEG 2000 Main Header syntax.

Figure 6 - JPEG 2000 First Tile-part header of a given Tile.

Figure 7 - JPEG 2000 Non First Tile-part header of a given Tile.

To give a better idea of when to use each marker, the name of marker to be used is indicated in Table 2 in relation with the type of information (and role) that can be provided within a JPEG 2000 codestream.

2.3 Packet Syntax and layers organisation

Having established the composition of the headers and role of each segment, let now consider the composition of the data parts. As illustrated by Figure 8, the data parts consists of an aggregation of packets, where a packet is defined as a part of the bit-stream comprising a packet header and the compressed image data from one layer of one precinct of one resolution level of one tile-component. The order in which the packets are found is called the progression order.

Figure 8 - Organisation of data parts in a JPEG 2000 codestream.

The syntax of a packet is described in Figure 9, where one sees that the packet header can be (in error resilience mode) made protected by the presence of two markers: SOP and EPH delimiting the boundaries of the packet header. The packet body contains then the entropy coded data which is placed after the corresponding packet header. The packet header itself has a precise syntax, relying on the definition of a Tag Tree. It is not the purpose here to detail how to compute the Packet header and the Tag Tree, but any interested user can refer him/herself to [1] for more information.

Figure 9 - Syntax of a JPEG 2000 packet.

3. Dependencies inside the JPEG 2000 image Coding Format

Contrarily to predictive coding standards such as MPEG-x and H.26x ones, M-JPEG 2000 bitstreams are not organised as inter-dependant frames. Using the traditional notations of predictive coding solutions, a M-JPEG 2000 codestream is then made only of Intra frames (I) which are non other than JPEG 2000 pictures, as illustrated by Figure 10.

Figure 10 - M-JPEG 2000 : a bitstream made of Intra frames.

Having stated that, it is easy to understand that the dependencies that will be observed in M-JPEG 2000 video streams are also dependencies present in JPEG 2000 codestreams. In particular, one understands easily that scalability features, relying on layered dispositions will lead to such dependencies. The reader interested in full details of the standard can refer him/herself to the complete standard [1]. We will here give an example of layers information repartition, to illustrate the principle of layers used in JPEG 2000. From the organisation of data packets presented in Figure 8, it is easy to understand the dependencies between layers and subbands. Typically, when a packet is lost, depending on the progression order, packets from layers greater than l, or componants greater than m in its subband n or packets from subbands greater than n can be lost.

Moreover, dependencies can result from the progression order in which the packets are transmitted depends of the organisation of the codestream. In the case of a progressive quality layer bitstream organisation, one will see the packet progression order providing the packets layer after layer, as illustrated by Figure 11. In the case of a progressive resolution bitstream resolution, as illustrated by Figure 12, the packets will be ordered so as to have the resolution refinements placed one after the other.

Figure 11 - Progressive quality layer bit-stream organisation.

Figure 12 - Progressive resolution bit-stream organisation.

4. Reference