Recent advances in data compression

Image and image sequence contents often represent a substantial volume of data. In most cases, there is a statistical redundancy among samples of such data that if appropriately modeled, can reduce their volume. Also, as the vast majority of image content is destined to be consumed by human subjects, further compression can be obtained by taking into account the properties of the human visual system.  Practical considerations such as complexity, memory requirements, delay, progressiveness, backward compatibility, lossy or lossless transcoding, error resiliency, etc. often play an important role when designing and deploying useful image and video compression technologies. 
In this presentation, we start with an overview of the design choices that led to the popular JPEG image compression format with emphasis on its advantages and drawbacks. We then discuss more recent image compression formats such as JPEG 2000, JPEG XR, JPEG XS, JPEG XT, and JPEG XL, along with motivations that led to their design and the added values they bring. The recent JPEG Pleno compression standard that allows efficient compression of new imaging modalities such as lightfield, point cloud, and holography is then discussed. Last but not least, the role of machine learning and in particular the impact of the paradigm shift from humans as consumers of images and video, to machines and computer vision systems, will be discussed, and emerging ideas on how to design image compression based on deep learning will be presented.