JPEG AI: The First International Standard for Image Coding Based on an End-to-End Learning-Based Approach
Alshina, E.
;
Ascenso, J.
; Ebrahimi, T.E.
IEEE Multimedia Vol. 31, Nº 4, pp. 60 - 69, October, 2024.
ISSN (print): 1070-986X
ISSN (online): 1941-0166
Scimago Journal Ranking: 0,68 (in 2024)
Digital Object Identifier: 10.1109/MMUL.2024.3485255
Abstract
JPEG AI represents a groundbreaking achievement as the first international standard for end-to-end learning-based image coding. Developed by the JPEG Standardization Committee, this innovative approach utilizes deep neural networks to achieve superior rate-distortion performance, offering enhanced perceptual visual quality and faster coding capabilities. The codec supports a wide array of devices, including mobile platforms, through optimized encoding and decoding processes. JPEG AI’s architecture enables multi-purpose optimization for both human visualization and machine-driven tasks, fostering efficient image processing. By leveraging GPUs and NPUs, it significantly reduces encoding and decoding times compared
to traditional codecs, such as HEVC/H.265 and VVC/H.266, while achieving higher compression efficiency.
Expected to become an official standard in early 2025, JPEG AI paves the way for advanced image compression with future extensions aimed at enhancing computer vision and image processing capabilities.
