3 ms·
No. The error image just contains "the hard stuff".
by zellyn 15y ago
No. The error image just contains "the hard stuff".
- moultano 15y agoThat's not really true. The residuals of a good lossy compression algorithm ought to look like uncorrelated noise with a smaller dynamic range than the original image, and the second fact (smaller dynamic range) makes them compressible.
- DarkShikari 15y agoDynamic range is not really relevant to compression. Entropy is much more meaningful, and error images typically have a huge amount of entropy.
- moultano 15y agoThat doesn't make any sense. You've taken an image with arbitrary bytes, and turned it into one where the bytes are tightly centered around a small range of values.That's perfect for Golomb coding (for instance.)
- jerf 15y agoNo free lunches. For a lossless compressor, (Information theory bits of polygons) + (Information theory bits of remaining error) >= (Information theory bits of original image), where the > represents the possibility that the first two elements aren't perfectly separated by your process or have inescapable overlap. I specify "information theory bits" because they aren't really what you see in the computer's RAM; they're closer to "post-compression bits". But regardless, no matter how you move the encoding around there is no escaping information theory.
- moultano 15y agoObviously. That's the definition of compression. If the image is well modeled by overlapping polygons + small residuals, the encoding will better approach the uncomputable ideal, and thus, compress.