In Sydney 86th JPEG meeting. Among other activities, the JPEG Committee issued (CfE), which is aimed at developers. The fact is that a year ago, the committee's specialists began research on the use of AI for image encoding. They, in particular, had to prove the advantages of neural networks over traditional methods.
As part of the JPEG AI initiative, it is supposed to improve the efficiency of image compression, but the downside is the need to train neural networks on large amounts of data. The Call for Evidence (CfE) was published following a meeting jointly with IEEE ICIP 2020.
In addition, within the framework of the JPEG Pleno system, work is underway to integrate various forms of plenoptic content into a single structure for processing it in a seamless mode. This technology is based on the vector field of light rays created by the lens, while classical lenses use the effect of illumination distribution in the plane of the actual image.
The JPEG Committee believes that to improve the performance of JPEG Pleno, cloud processing of such images should be added, which will speed up the process and improve the final result. After all, the JPEG standard has been around for many years, and technologies are developing, therefore, what is already there should be improved.
It has not yet been announced when the use of neural networks for image encoding and cloud processing will become industry standards, but the first steps in this direction have already been taken.
Source: 3dnews.ru