Highly Flexible Image Coding Using Non-linear Representations

This paper presents a new image representation method based on a Matching Pursuit expansion over a dictionary built on anisotropically refined atoms. New breakthroughs in image coding certainly rely on truly multidimensional signal decompositions, and the coder proposed in this paper provides an adaptive way of representing images as a sum of two-dimensional features. It is shown to provide very competitive results with the state of the art in image compression, as represented by JPEG2000. The visual quality also clearly favors the Matching Pursuit scheme, whose coding artifacts are less annoying than the ringing introduced by wavelets at very low bit rate. In addition to good compression performance at low bit rate, the new coder has the great advantage of producing a high flexibility scalable bitstream, which is becoming a very important feature in today’s visual communication applications. The Matching Pursuit stream can easily be decoded at any spatial resolution, different from the original image, and the bitstream can very simply be truncated at any point to match diverse bandwidth requirements. These spatial and rate scalability features are shown to be way more flexible and less complex than transcoding operations generally applied to state of the art streams. Thanks to both its capacity for efficient representation of multidimensional signals, and its very flexible structure, the novel image coder proposed in this paper certainly opens interesting new perspectives in image coding and compression for visual communication services.