JPEG Column: 89th JPEG Meeting

Author: Antonio Pinheiro
Affiliation: Instituto de Telecomunicacoes (IT) and Universidade da Beira Interior (UBI), Portugal

JPEG initiates standardisation of image compression based on AI

The 89th JPEG meeting was held online from 5 to 9 October 2020.

During this meeting multiple JPEG standardisation activities and explorations were discussed and progressed. Notably, the call for evidence on learning-based image coding was successfully completed and evidence was found that this technology promises several new functionalities while offering at the same time superior compression efficiency, beyond the state of the art. A new work item, JPEG AI, that will use learning-based image coding as core technology has been proposed, enlarging the already wide families of JPEG standards.

Figure 1. JPEG Families of standards and JPEG AI.

The 89th JPEG meeting had the following highlights:

  • JPEG AI call for evidence report
  • JPEG explores standardization needs to address fake media
  • JPEG Pleno Point Cloud Coding reviews status of the call for evidence
  • JPEG Pleno Holography call for proposals timeline
  • JPEG DNA identifies use cases and requirements
  • JPEG XL standard defines the final specification
  • JPEG Systems JLINK reaches committee draft stage
  • JPEG XS 2nd Edition Parts 1, 2 and 3.

JPEG AI

At the 89th meeting the submissions to the Call for Evidence on learning-based image coding were presented and discussed. Four submissions were received in response to the Call for Evidence. The results of the subjective evaluation of the submissions to the Call for Evidence were reported and discussed in detail by experts. It was agreed that there is strong evidence that learning-based image coding solutions can outperform the already defined anchors in terms of compression efficiency, when compared to state-of-the-art conventional image coding architecture. Thus, it was decided to create a new standardisation activity for a JPEG AI on learning-based image coding system, that applies machine learning tools to achieve substantially better compression efficiency compared to current image coding systems, while offering unique features desirable for an efficient distribution and consumption of images. This type of approach should allow to obtain an efficient compressed domain representation not only for visualisation, but also for machine learning based image processing and computer vision. JPEG AI releases to the public the results of the objective and subjective evaluations as well as a first version of common test conditions for assessing the performance of leaning-based image coding systems.

JPEG explores standardization needs to address fake media

Recent advances in media modification, particularly deep learning-based approaches, can produce near realistic media content that is almost indistinguishable from authentic content. These developments open opportunities for production of new types of media contents that are useful for many creative industries but also increase risks of spread of maliciously modified content (e.g., ‘deepfake’) leading to social unrest, spreading of rumours or encouragement of hate crimes. The JPEG Committee is interested in exploring if a JPEG standard can facilitate a secure and reliable annotation of media modifications, both in good faith and malicious usage scenarios. 

The JPEG is currently discussing with stakeholders from academia, industry and other organisations to explore the use cases that will define a roadmap to identify the requirements leading to a potential standard. The Committee has received significant interest and has released a public document outlining the context, use cases and requirements. JPEG invites experts and technology users to actively participate in this activity and attend a workshop, to be held online in December 2020. Details on the activities of JPEG in this area can be found on the JPEG.org website. Interested parties are notably encouraged to register to the mailing list of the ad hoc group that has been set up to facilitate the discussions and coordination on this topic.

JPEG Pleno Point Cloud Coding

JPEG Pleno is working towards the integration of various modalities of plenoptic content under a single and seamless framework. Efficient and powerful point cloud representation is a key feature within this vision. Point cloud data supports a wide range of applications including computer-aided manufacturing, entertainment, cultural heritage preservation, scientific research and advanced sensing and analysis. During the 89th JPEG meeting, the JPEG Committee reviewed expressions of interest in the Final Call for Evidence on JPEG Pleno Point Cloud Coding. This Call for Evidence focuses specifically on point cloud coding solutions supporting scalability and random access of decoded point clouds. Between its 89th and 90th meetings, the JPEG Committee will be actively promoting this activity and collecting submissions to participate in the Call for Evidence.

JPEG Pleno Holography

At the 89th meeting, the JPEG Committee released an updated draft of the Call for Proposals for JPEG Pleno Holography. A final Call for Proposals on JPEG Pleno Holography will be released in April 2021. JPEG Pleno Holography is seeking for compression solutions of holographic content. The scope of the activity is quite large and addresses diverse use cases such as holographic microscopy and tomography, but also holographic displays and printing. Current activities are centred around refining the objective and subjective quality assessment procedures. Interested parties are already invited at this stage to participate in these activities.

JPEG DNA

JPEG standards are used in storage and archival of digital pictures. This puts the JPEG Committee in a good position to address the challenges of DNA-based storage by proposing an efficient image coding format to create artificial DNA molecules. JPEG DNA has been established as an exploration activity within the JPEG Committee to study use cases, to identify requirements and to assess the state of the art in DNA storage for the purpose of image archival using DNA in order to launch a standardization activity. To this end, a first workshop was organised on 30 September 2020. Presentations made at the workshop are available from the following URL:

http://ds.jpeg.org/proceedings/JPEG_DNA_1st_Workshop_Proceedings.zip.

At its 89th meeting, the JPEG Committee released a second version of a public document that describes its findings regarding storage of digital images using artificial DNA. In this framework, JPEG DNA ad hoc group was re-conducted in order to continue its activities to further refine the above-mentioned document and to organise a second workshop. Interested parties are invited to join this activity by participating in the AHG through the following URL: http://listregistration.jpeg.org.

JPEG XL

Final technical comments by national bodies have been addressed and incorporated into the JPEG XL specification (ISO/IEC 18181-1) and the reference implementation. A draft FDIS study text has been prepared and final validation experiments are planned.

JPEG Systems

The JLINK (ISO/IEC 19566-7) standard has reached committee draft stage and will be made public.  The JPEG Committee invites technical feedback on the document which is available on the JPEG website.  Development of the JPEG Snack (IS0/IEC 19566-8) standard has begun to support the defined use cases and requirements.  Interested parties can subscribe to the mailing list of the JPEG Systems AHG in order to contribute to the above activities.

JPEG XS

The JPEG committee is finalizing its work on the 2nd Editions of JPEG-XS Part 1, Part 2 and Part 3. Part 1 defines new coding tools required to efficiently compress raw Bayer images. The observed quality gains of raw Bayer compression over compressing in the RGB domain can be as high as 5dB PSNR. Moreover, the second edition adds support for mathematically lossless image compression and allows compression of 4:2:0 sub-sampled images. Part 2 defines new profiles for such content. With the support for low-complexity high quality compression of raw Bayer (or Color-Filtered Array) data, JPEG XS proves to also be an excellent compression scheme in the professional and consumer digital camera market, as well as in the machine vision and automotive industry.

Final Quote

“JPEG AI will be a new work item completing the collection of JPEG standards. JPEG AI relies on artificial intelligence to compress images. This standard not only will offer superior compression efficiency beyond the current state of the art but also will open new possibilities for vision tasks by machines and computational imaging for humans.” Said Prof. Touradj Ebrahimi, the Convenor of the JPEG Committee.

Future JPEG meetings are planned as follows:

  • No 90, will be held online from January 18 to 22, 2021.
  • N0 91, will be held online from April 19 to 23, 2021.
Bookmark the permalink.