SAN FRANCISCO — Condensation, a perfectly lossless compression technology derived from quantum thermodynamics, achieved a new high in image compression this week when BitJazz Inc. announced a 2.5 average compression ratio for PhotoJazz.

A free decoder, available from the BitJazz Web site, enables images compressed into the PhotoJazz format to be fully decompressed at ratios ranging from 1.92 to 2.98 on Kodak's standard suite of test images. A software development kit enables any engineer to incorporate the new compression method into existing and new applications.

BitJazz calls its approach "condensation" instead of compression, because of its roots in information theory. That discipline came out of quantum thermodynamics theory, also known as statistical mechanics.

According to BitJazz, the notion of entropy originated with statistical mechanics, and was first applied by Claude Shannon, the founder of information theory, when he noticed that negative entropy and information were identical.

"We have taken the analogy between negative entropy and information further than Shannon in our application to compressing image data — that's what we call information condensation," said Andreas Wittenstein, founder of BitJazz (San Francisco) and inventor of the technology used in PhotoJazz. Wittenstein and company vice president Jim Lloyd previously developed Pixar's Toy Story CD-ROMs and Quick RenderMan, as well as Speech Systems' Phonetic Engine.

"Information is mathematically identical to negative entropy," Wittenstein said, "and in a sense entropy is the measure of how much information there is in the data set. If any possible pixel was likely to be next to any other — say in a white-noise image — then there would be no information present. But in real images the probabilities of which pixel is going to be next to which other contains the information."

What PhotoJazz does, Wittenstein said, is to "find these patterns in image data and discover an implicit organization that is not immediately obvious otherwise."

PhotoJazz supports the full repertoire of color modes, including RGB, CMYK, gray scale, duotone, Lab and multichannel image data. Thumbnails, profiles and other auxiliary image data are also preserved. Essentially, the company claims, any type of image information can be condensed in a perfectly lossless manner.

"We call it perfectly lossless, because you get back bit for bit what you started with by taking advantage of the correlation among pixels that always exists in image data," said Lloyd. "Most other approaches work a scan line at a time, but PhotoJazz achieves its 2.5 average compression ratio by taking advantage of both vertical and horizontal correlations."

A software development kit (SDK) enables developers to license the PhotoJazz algorithms for incorporation into applications running under either Windows or the MacOS. The basic developer's license allows use of the algorithms free of charge; users can read PhotoJazz images free, too.