At this aim, the Versatile Video Coding (VVC) standard has recently been finalized (Bross et al., 2021), achieving approximately another 50% bit rate reduction for the same subjective quality when compared to HEVC. However, the adoption of new standards is very limited in most imaging applications; there are several motivations, one reason is that these standards are ruled by several patent pools with different pricing structures and terms and conditions; another reason is the fact that very often, in real application scenarios the available computing power is not sufficient for an effective decoding process; in addition, in many cases there are concerns about backward compatibility with previous versions of the codecs.
So, there is still plenty of space for the adoption of those recent standards, applications. A similar situation holds for the case of image coding standars: here, JPEG (Hudson et al., 2017) is currently the Spain phone number list most widely used lossy image compression standard, although it was introduced in 1992. Its successors, starting from JPEG 2000 (Taubman and Marcellin, 2002), to the High Efficiency Image File Format (HEIF) (Lainema et al., 2016), and to the AV1 Image File Format (AVIF), which is the latest image compression standard, have not been sufficiently spread in imaging applications.
Robustness of Deep Learning-Based Solutions Imaging applications have all seen the technological migration from model-based algorithms to data driven-based algorithms. A model-based approach, relying on some mathematical or statistical models of the application scenario and the involved data, has usually good performance, provided that the application follows the designed model, whereas it has an in most cases a smooth degradation of performance if this model does not correctly represent the real world scenario. More recent methods, instead, are for the most part data-driven: they exploit the availability of large amounts of data to learn how to solve the problem at hand.