21.5. PNG Compression
The most notable aspect of PNG compression is that it is "lossless," meaning no information is lost in the compression process. A decompressed PNG image is identical to the original.
PNGs use a "deflate" compression scheme (the same engine used to "zip" files with gzip, WinZip, etc.). Like GIFs, PNG's compression works on rows of pixels, taking advantage of repetition in bytes of information. By use of internal filters, it can take advantage of some vertical patterns as well. PNG's compression engine typically compresses images 5-25% better than GIF (and up to 39% better under optimal conditions). Unfortunately, with the currently available tools, it is difficult to achieve such file savings. See Section 21.7, "Creating PNG Files" later in this chapter.
21.5.1. Filters
Before PNG compresses an image, it first runs the image data, row by row, through one of five filters (Sub, Up, Average, Paeth, Adaptive). The filters use different methods for finding patterns in the image information that can then be condensed more efficiently. The process is similar to how LZW compression takes advantage of horizontal repetition in GIFs, but PNG can look for vertical repetition as well.
In most applications, the filters are applied internally and are hidden from the end user (as they should be). If your tool provides filter options, there are only two you need to remember:
-
Use "None" for all indexed color images (or grayscale images with fewer than 16 shades).
-
Use "Adaptive" for all other image types.
 |  |  |
| 21.4. 8-Bit Palette, Grayscale, and Truecolor |  | 21.6. Special Features |
Copyright © 2002 O'Reilly & Associates. All rights reserved.