Performance Bounds on Image and Video Compression

当前多媒体应用和服务的激增主要是由图像和视频压缩技术的重大最新进展实现的，JPEG用于静态图像，MPEG用于视频，这已经成为家庭和营销术语。 数字HDTV即将出现，它将以MPEG-2标准为基础。 这是否意味着图像/视频编码已经达到饱和状态，在这种状态下，对压缩的更多研究不太可能产生显着的改进？ 本研究试图通过研究图像和视频压缩的基本性能界限来阐明这一点。 一个主要的目标是发现现有的商业系统和理论性能之间的差距，最佳可达到的，并在未来指导改进的压缩算法的生成。 本研究的主要重点是获得基本的率失真界限的现实类的图像和视频模型。 这使得能够生成对这些模型理论上可达到的最佳性能与基于它们的特定编码算法的最佳性能之间的差距的可靠估计。 本研究的分析方法包含以下关键组成部分： (1)越来越复杂和现实的图像/视频模型的层次结构;（2）这些模型的率失真界限的推导;（3）模型参数可识别性和估计精度的分析;（4）基于通用谱盲时空采样技术的压缩;以及（5）基于高性能实用算法的真实的图像/视频的模型验证。

The current explosion in multimedia applications and services has been primarily enabled by the significant recent advances in image and video compression technology, e.g., JPEG for still images and MPEG for video, which have become household and marketing terms. Digital HDTV is on the horizon, and will be based on the MPEG-2 standard. Does this mean that image/video coding has reached a state of saturation where more research on compression is unlikely to yield significant improvements? This research tries to shed light on this by investigating the fundamental performance bounds on image and video compression. A primary goal is to uncover the performance gaps between existing commercial systems and the theoretical performance optimally attainable, and to guide the generation of improved compression algorithms in the future. The main focus of this research is to derive fundamental rate--distortion bounds for realistic classes of image and video models. This enables the generation of reliable estimates of the gap between the optimal performance theoretically attainable for these models and that of specific coding algorithms based on them. The analytical approach of this research contains the following key components: (1) a hierarchy of increasingly complex and realistic image/video models; (2) derivation of rate-distortion bounds for these models; (3) analysis of model parameter identifiability and estimation accuracy; (4) compression based on universal spectrum-blind spatio-temporal sampling techniques, and (5) model validation on real images/video based on high-performance practical algorithms.