Giovambattista Pulcini's
pages on Fractal Image Compression

Special thanks to Valerio Verrando for hosting this page!

IFSAF Hardware and Software requirements

• You should use at least a 486DX2-66 with 8Mb RAM, a SVGA with 15 or 16 bpp (this is only for best viewing the results, since IFSAF doesn't handle dithering on 8bpp displays, anyway you may view them with another program as LView or PSP) Windows 3.10 or higher running in Enhanced mode with a 15Mb swap file and some available space on the HD: say 2-3 Mb at least.
  The program has been reported to run under Windows 95 too.
  These are not absolute minimal requirements, but those for a reasonable working configuration for encoding images 256x256 or 512x512, depending on options settings.

• All files relating to IFSAF must be placed in the toplevel directory: \IFS or it won't be able to find some of them. Put here the images too.

• The input image must be a .BMP windows file in true-color RGB format, not RLE compressed, which is not supported. The dimensions of the image must be power of two or at least divisible by the size of range blocks: they must contain an integer number of range blocks.
  For gray-level images, .BMP RGB encoded 8bpp with palette, we take the pixel color index as the level of gray, not the value indexed in the palette.

• No other program should be running while IFSAF is running. You should keep free as much RAM as you can.

• IFSAF is not a well behaved windows program: during encoding/decoding it monopolizes the CPU and doesn't release it until the encoding is complete, preventing other programs to run.
  There is no way to stop the encoding or decoding once started, besides you can kill the program by pressing Ctrl-Alt-Del once and then Return.

• Use image at most of 512x512 (with 8Mb RAM) or you will get your system trashing during encoding. Some swap during decoding is ok.

• With parameters at extreme values, you may get some sqrt:DOMAIN errors.

• Sometimes it's better to close the status window bebore you start the encoding or the decoding, since program may hang, in some cases.

Actually some features available for gray-scale image compression are not available for true color image compression:

• Local Sub Search (LSS) and any related option.
• Two levels quadtree partitioning.
• Polynomial approximation of shade blocks.
• Simulated annealing encoding.

The IFSAF underlying theory.

---

Giovambattista Pulcini

Last Revised: April, 1997 - ver. 2.01