Why Tiny JPEGs Look Different in Chrome

Why Tiny JPEGs Look Different in Chrome(guillaumetech.github.io)

98 points by gutechh 2 hours ago | 17 comments

I'm pretty sure the same issue happens with PNGs, which being lossless are generally good for icons as they don't land up with compression artifacts like what happens in JPEGs (which the author points out are really for photographs), they also support alpha blending.

When Chrome introduced this "optimization" and it made it through to an Electron release which we were upgrading to, it really messed up the icons in a lot of places in our product such that we had to hold off the upgrade until we had SVGs to replace them.

SVGs also have the advantage of being able to respond to light/dark mode. We just needed to put each icon in its own shadow DOM to avoid styles clashing between the different SVGs if they happen to be named the same which was a bit of an annoyance for our graphic designer.

gutechh 7 minutes ago | |

Ah this is weird! My understanding is that png doesn't store the images like jpg at all, they use a palette of all the colors in the image. Then each pixel is a reference to this palette. So it's not IDCT scaling, but I imagine there are other tricks to save on space/cpu when decoding pngs.

StilesCrisis 28 minutes ago | |

You were using bitmap icons that were more than 8x larger on each axis than necessary? Why?

aljgz 23 minutes ago | | |

Guess: same icon used in very small size for lists/menus and much larger when you select the item. Also, if the same one is used for all resolutions, the actual size can vary widely between monitor sizes, DPIs, resolutions, scales. Good to have one that looks decent to people who see it large and/or care about quality

muizelaar 1 hour ago |

The work for decompressing at a lower scale in Firefox is happening here: https://bugzilla.mozilla.org/show_bug.cgi?id=2033250

gutechh 58 minutes ago | |

Thanks, I knew it was much more efficient, but seeing the numbers in the benchmarks make me appreciate this "trick" even more!

debazel 1 hour ago |

Chrome and Firefox uses two different scaling algorithms that is probably contributing a lot more to this difference. Chrome is more blurry in general while Firefox is sharper but has slightly more ringing artifacts. Personally I prefer the Firefox version.

aidenn0 1 hour ago | |

I've noticed that a lot of objective measures of image quality prefer blur to ringing, but many subjective measurements tilt the other way.

ryandamm 1 hour ago | | |

That's why you should prefer LPIPS to PSNR. LPIPS is a learned metric that uses subjective human input.

That said, it's limited due to the training set, iirc. So we still use PSNR regularly — it's an easily-understood metric with widely-understood limitations, which is in practice not too bad. Devil you know, etc.

gutechh 1 hour ago | |

I do prefer the firefox look too! I'd be curious to hear your take on what could be the root cause of this. I'm far from a image rendering expert. But I remember when inspecting the edges and curves of the images in a DCT visualizer they were all in the AC coefficients which is what led me to the conclusion of the post.

debazel 1 minute ago | | |

I'm far from an image rendering expert myself, but I don't think there is a single root cause here. I don't doubt the partial decoding would affect the final result just as the scaling algorithm itself, so in the end it's going to be a combination of it all.

The reason I believe the scaling algorithm is contributing more in your particular example, is because it looks almost exactly like a classic B/W comparison between a sharp ringing heavy algorithm and a more blurry one.

For reference here is a quick comparison between Lanczos/Bilinear I threw together: https://twinlens.app/compare?share=40973d84a174

It would be interesting to see how your example would look if you used colors instead, where the ringing artifacts become a lot more obvious. Like this one: https://twinlens.app/compare?share=8fb890a5820e

ack_complete 50 minutes ago | | |

I wonder if this is a gamma correction related. JPEG encodes directly in a non linear color encoding (full-range YCbCr), so the partial decoding may be effectively scaling without gamma correction.

PetitPrince 1 hour ago |

So is Firefox doing a full rendering then scaling, or is it also doing a partial rendering but in another way? You're only telling one side of the story.

the8472 1 hour ago | |

Firefox doesn't decode the image into a full-resolution buffer and then downscale either, instead it does a streaming decode and applies downscaling on the fly. They call it downscale-during-decode[0]. I guess it doesn't do partial decoding of the 8x8 blocks though.

[0] https://bugzilla.mozilla.org/show_bug.cgi?id=1045926

gutechh 1 hour ago | |

Ah! It took a fair bit of digging to get there. Maybe I'll look into the firefox pipeline and make another post. In the mean time if anyone know the internals and how firefox does it, I'm eager to hear about it!

bluedino 1 hour ago |

Reminds me of digging into HTML Canvas scaling, especially when HiDPI Macs came out.

smallnix 1 hour ago |

Would be nifty if they only did that partial scaling when there's little memory available.

gutechh 1 hour ago | |

ahah, yeah surprising considering how Chrome is RAM hungry, 20mb isn't that much of a win. On another end when trying to reproduce the glitch, it was quite hard! basically invisible on photos and not all icons were affected. So it does work very well. It was kind of our fault to use a jpg for an icon ahah.

LoganDark 37 minutes ago |

Not sure why one would use JPEG for a little digital icon anyway. It is indeed only acceptable for photos. SVG is a good choice for icons, PNG or WebP for other digital content.

gutechh 18 minutes ago | |

Yes! for a bit of backstory, it was an internal icon on a prototype that we were polishing up. So I get why the person did it at the time, running after the svg would have been too much effort for something that might have been tossed in the end.