explain the degradation of quality

Jpeg

Jpeg

File: needs_more_jpg_king_of_the_hill.webm (365KB, 480x360px) Image search: [Google]

365KB, 480x360px

>>258508
>>258510

https://youtu.be/QEzhxP-pdos

DCT coefficients are used to construct an image, and it takes a while to guess which ones make up an image already compressed with DCT, so it guess enough to get a certain amount of similarity, and, thus, the quality will ultimately degrade as the image is saved and reuploaded to media outlets

>>258504
Jpeg is lossy. When you use it to save an image, it stores it in a way that optimizes similarity to the original image to a degree, this similarity can be linked to the quality of it. When you transform the compressed data back into spatial domain, the result is similar, but not quite the same. The difference between the input and the output can be thought of as the error.
Now, when you compress again by using the compressed image as the source, the encoder tries to fit the data to the already noisy image, and the error (=difference between input and output) stacks up.
The blocky appearance is due to how jpeg works - he image is converted into Y'CbCr color space - gamma-corrected luma (Y'), and two chrominance channels (Cb and Cr). Chrominance is then downsampled further (usually using 4:2:0 subsampling, which means the resolution is halved both horizontally and vertically), and then split into blocks, most commonly 16x16 pixels. and transformed into a linear combination of DCT basis functions (which look like gradients in x/y directions), and quantized according to set quality value. And when you transform them back, some blocks are not reproduced exactly and you get discontinuities at the edges which manifests as visible blocks.

>>258532
So do social media sites like Facebook, twitter, etc further compress images when you post them? I realize most of the time it's fucking retards screenshotting and cropping memes, but even if you optimize a png for web, the uploaded result is a bit blocky.

>>258504
>True font
>JPEG 1/10
Yep

>>258533
Yeah, probably. I mean, they have to pay the bills, and less data transferred is less money wasted.
They also probably prepare several resolution/quality variants and don't give you the best by default. Twitter does this (:small, :large, :orig), tumblr does this (_1280, _500, ...), don't know about facebook but probably too. Imgur recompresses if it's above a certain threshold, which is pretty much most of the time.

>>258533
I've noticed that Facebook messenger converts my memes to jpg, as does my Android when I send things over MMS.

>>258508
>>258510


It's called artifacting you dicklickers.

Identify the disease not the symptom.

>>258533
Can't really speak to everything else but Facebook's compression in particular is atrocious, especially with reds. Try upload a plain red square on a white background or something, I guarantee it'll end up looking like complete trash.

>>258606
Maybe they subsample chroma differently.

>>258532
>Jpeg is lossy
Everything you are saying is correct except for this. a .jpg CAN be lossy, it's dependent on the initial rendering/formatting.

Some software goes automatic mode and compresses, other software allows you to render at %100 with TIFF comparable color palette range and object hinting comparable to a .png

I'll probably catch some flak for this, but this is just one of many areas where AI fails, AI handles rendering vectors so poorly.

>>258611
>.jpg CAN be lossy
Well, there are standards which include lossless encoding, like JPEG2000. But with the original JPEGs, wouldn't the quantization step and rounding to the nearest integer always lose some data?

>AI handles rendering vectors so poorly
What do you mean by that?

>>258630
Yes sorry I wasn't more clear, the 2000 format will get you close to lossless bitmap- less postscript data than a TIFF but for reproduction almost identical. A standard jpg, uncompressed, with 26 million color bits, object hinting, and anti aliasing is basically indistinguishable from the 2000 format basically a wash - henceforth why the 2000 format is not really used.

So AI has 3 big rendering flaws.

1. You cannot scale the graphic when exporting - might seem silly, but often professionally you build you work very proportionally, needing to duplicate, make a new artboard, and possibly change the measurement tools to get the right exported graphic is lame and clunky. And at times, for large format reproduction I've had to open the AI file in PS because however their shit code is written - AI hangs up on rendering large format bit maps.

2. Color correction. AI does this shit where you turn proofing colors on - but they don't' change shit. So then you take your print ready pantone matched art and export it as a RGB jpg, the display looks the same - because AI has shit proofing display, but your art will now reproduce totally different in print. It's a huge oversight and can sabotage you as a designer.

3. because of the exporter style, you cannot switch measurements to inches, millimeters, pixels or points during the exporting process. This forces you to not just work in scale ( which we all do), but to define or change your document's measurement settings based off of a simple action like exporting a rendered graphic.

File: renderstuff.jpg (509KB, 1728x775px) Image search: [Google]

509KB, 1728x775px

>>258630

>>258632
Shit, nigga, I thought you meant artificial intelligence for some reason. My bad.

File: 1294601401031.jpg (5KB, 192x197px) Image search: [Google]

5KB, 192x197px

>>258632
>Antia-aliasing : Type Optimized (Hinted)
I'll stick with Inkscape then.

>>258641
Yep cause clearly that screencap is me exporting a finished file, and not a representation of file options.