A novel content-adaptive image compression system

This paper presents a novel content-adaptive image compression system. Utilizing a pattern-driven model, we explore the synergy between content-based analysis and compression. For a given image, disparate low-level visual patterns are automatically separated, modeled, and encoded using compact and "customized" features and parameters. The feasibility and efficiency of the proposed system were corroborated by quantitative experiments and comparisons. Since different patterns are separated and modeled explicitly during the compression, our method holds potentials for providing better support for compressed-domain analysis.