Fractal image compression developed from pure mathematics in the 1980s and became a commercial technology in the 1990s. This page describes its development and why it did not become a standard.
The word fractal was introduced by the mathematician Benoit Mandelbrot in 1975. In The Fractal Geometry of Nature (1982) he argued that coastlines, clouds, mountains and plants are better described by shapes that repeat their structure at every scale than by the lines and circles of classical geometry. The book made the idea widely known.
The mathematical tool that fractal compression uses came from a different direction. In 1981 John Hutchinson showed that a shape made of several reduced copies of itself is fully determined by the reducing maps alone: apply the maps repeatedly to any starting shape, and the result converges to one and the same limit. Michael Barnsley and Stephen Demko named such a set of maps an iterated function system in 1985 and showed how it can generate complex natural forms from a few numbers. Barnsley's fern, drawn by four affine maps, became the standard example.
For image compression, the problem is the reverse: to find the maps for a picture that already exists. In 1986 Barnsley, Vincent Ervin, Douglas Hardin and John Lancaster proved the collage theorem, which says that maps that approximately reproduce a picture from pieces of itself also have an attractor close to that picture. The encoder therefore only has to search for good approximations, piece by piece.
Barnsley led a research group on the subject at the Georgia Institute of Technology from 1985, with funding from the US defence research agency DARPA. In 1987 he and his colleague Alan Sloan founded Iterated Systems, Inc. to commercialise the work. In January 1988 they described the approach to a wide audience in the magazine BYTE, under the title “A Better Way to Compress Images”.
At that stage the method still lacked a general, automatic way to find the maps for an arbitrary photograph. That step came from Arnaud Jacquin, Barnsley's doctoral student at Georgia Tech. In his thesis of 1989, and in papers in 1990 and 1992, he described what became known as a partitioned iterated function system. Instead of looking for maps that describe the whole picture at once, Jacquin divided the picture into small blocks and approximated each block by a larger block from elsewhere in the same picture. This block-based method is the basis of practically every fractal image codec since, including FIF2.
Iterated Systems held more than twenty patents on the technology and sold it in several forms: an encoder called Fractal Imager, a stand-alone viewer, a plug-in for the Netscape browser and a developer kit for Windows. Its images used the file format FIF, for Fractal Image Format. The best-known user was Microsoft, which licensed the technology for the Encarta multimedia encyclopedia and used it to fit a large number of photographs onto a CD-ROM.
The company also developed fractal video codecs. SoftVideo was licensed to the games publisher Spectrum HoloByte in 1994, and ClearVideo was supported for years by Windows Media Player. In 1997 the Altamira Group released Genuine Fractals, a Photoshop plug-in licensed from Iterated Systems that could save pictures as FIF and was used mainly to enlarge photographs for print. The product passed to LizardTech in 2001 and to onOne Software in 2005, and was later sold as Perfect Resize and ON1 Resize.
Two books from this period remain the standard references: Fractal Image Compression by Michael Barnsley and Lyman Hurd (1993), and Fractal Image Compression: Theory and Application, edited by Yuval Fisher (1995).
Several factors worked against fractal compression:
In 2001 Iterated Systems renamed itself MediaBin and turned to other software; Interwoven acquired it in 2003. Academic research continued into the 2000s, mostly on faster encoding and on combinations with other methods, but fractal compression disappeared from everyday use.
©2026 By Oskar Behzadpour & Rezar Behzad | FIF v2 Format created by Rezar Behzad