Machine learning (ML) has drastically pushed the frontier of every aspect of Artificial Intelligence (AI) in the past decade. It has been widely used in various computer vision, automation, image and video processing applications, leading to leapfrogging improvements in performance. Tremendous efforts have been dedicated to applying ML-based techniques to image and video coding, which can be roughly categorized into two groups: the first group can be termed as ML-based hybrid approaches, which incorporate machine-leaning techniques into traditional image and video coding systems, either as stand-alone pre- and post-processing modules, or as optimization techniques of the operation (e.g. parameter setting) of traditional image and video coding systems, or for designing modules (e.g. filters for in-loop deblocking or motion compensation) in traditional image and video coding systems. The second group are native ML- and AI-based algorithms that aim at replacing the traditional prediction with quantized transform coding framework with an end-to-end ML- and AI-based approach, e.g. by using auto-encoders in image and video coding. Many ML-based hybrid techniques have already been incorporated into conventional image and video coding systems, while some native ML- and AI-based algorithms have also showed promises and achieved respectable performances. Various standardization organizations including the JPEG, MPEG and JVET have also either already started defining new standards with ML and AI as the center of the coding system and/or key target application, or are starting to look into related technologies.
In this special issue of the EURASIP JIVP, leading researchers and practitioners in academia, industry and standard-bodies are invited to contribute and produce a go-to-reference of the state-of-the-art in ML- and AI-based image and video coding theories, algorithms, techniques, systems and standardization activities for the entire community. Submissions that focused on “brave new ideas” in the development of end-to-end ML- and AI-based image and video compression systems are especially encouraged, even if such schemes might still lag in performance as compared with highly optimized traditional approaches. In addition to compression, submissions related to image and video processing using AI and ML techniques are also welcome.
Prospective authors are encouraged to submit new, unpublished manuscripts for publication in the Special Issue. Subject of interests include, but are not limited to
- Machine learning based end to end image and video compression algorithms
- Image and video compression with/for AI applications
- AI-based content analysis, and generation
- Machine learning based parameters tuning and compression algorithm setting for legacy image and video compression standards, such as JPEG, AVC, HEVC, VVC and AV-1
- Machine learning based quality evaluation for image and video compression
- Machine learning based Quality of Experience (QoE) for end-to-end image and video acquisition and presentation
- Machine learning based end-to-end image and video compression systems
- Machine learning based image and video pre- and post-processing, spatial and temporal-super-resolution
- Implementation optimizations
- Large public and annotated test and training data sets with descriptions
Manuscripts will be due May 5, 2021, with first round of reviews made available to the authors by July 18, and final decision on August 30. Camera-ready files by September 13, 2021.
Lead Guest Editor
Dr. Jiangtao Wen, Tsinghua University, Beijing, China and Flora Lucis SA, Cran-Montana, Switzerland
Dr. Shan Liu, Tencent Media Lab
Dr. Moncef Gabbouj, Director, NSF IUCRC Center for Visual and Decision Informatics, Finland-Site, Professor, FIEEE
Department of Computing Sciences, Faculty of Information Technology and Communication Sciences, Tampere University
Dr. Mathias Wien, RWTH Aachen University, Germany