Hacker News new | past | comments | ask | show | jobs | submit login

Ultimately, 'real lossless' does not exist since all raster image data (or audio data) consists of quantized samples with a limited spatial (or temporal) resolution (and in case it was captured, as opposed to synthesized, there will be errors caused by the capturing process too). But as long as the precision is high enough to be indistinguishable by humans in the viewing conditions (or editing conditions) that matter to them, everything is fine. Whether that is achieved through lossless codecs (possibly used in near-lossless ways) or through lossy codecs is probably less important in practice.

What matters is getting the workflow right to minimize generation loss and precision loss during authoring. If your workflow critically depends on assumptions like "png is lossless, jpeg is lossy", you should probably re-think your workflow. A jpeg straight from a camera is 'more lossless' than a downscaled 8-bit png that was created from it. Just going by file type to determine where in the workflow you are, is not a good strategy.




> A jpeg straight from a camera is 'more lossless' than a downscaled 8-bit png that was created from it.

Well there's your problem right there: PNGs don't come from cameras. PNGs come from image generation software that produces specific, discrete pixel values. Changing #555555 to #565656 in lossless image data is no more acceptable than changing "UUU" to "VVV" in lossless text data. Unfortunately, gzip et al don't perform well on image data, so different but still lossless compression algorithms are required.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: