The Piecewise-Constant Image Model

Caravian's Piecewise-Constant Image Modeling technology is available for demonstration as a Microsoft ActiveX control for Internet Explorer. Our PWC Embedded Image Viewer should download automatically if you are viewing this page from a machine running Microsoft Windows 9x or NT. The image at the right is displayed using the Viewer.

PWC Examples

Vector Graphic Material

Since it is a true two-dimensional model, PWC performs best on highly structured synthetic material. Vector graphics such as  non-antialiased text, graphs, charts, maps, figures, metafiles, and user interface elements compress particularly well. For example, the following table compares GIF and PWC on some typical vector graphic material. On average, PWC compresses these images over four times better than GIF. Further, PWC is twice as fast on encode and 33% faster on decode. All times are given for a 200 MHz Intel Pentium (pre MMX) Processor .

Metric GIF PWC
Uncompressed bytes 981,324 981,324
Compressed bytes 56,972 13,558
Compression rate 17.2:1 72.4:1
Encode time (sec) 1.2 0.6
Decode time (sec) 0.9 0.6
Visual GIF PWC

Natural Images

Palettized natural images are produced by approximating each image color with a best representative from a reduced set of colors. The most sophisticated color reduction methods will locally average the color error through an error diffusion process. This process introduces a noise-like component that is difficult to compress losslessly. On images with the heaviest such dither, PWC typically compresses 25% better than GIF. On natural images that have been color reduced without the application of dither, PWC typically compresses 50% better than GIF.