On Mon Mar 24 12:22:57 2025 +0000, Matteo Bruni wrote:
Ah, so it wasn't chroma subsampling after all? :rolling_eyes: Did you see that from the JPEG metadata or just deduce it from the test results?
Yeah doesn't seem at all related to native using higher chroma subsampling like I suspected :/
I encoded a 4:4:4 image with libjpeg to confirm `identify -verbose` could actually identify a 4:4:4 JPEG, which resulted in: ``` jpeg:colorspace: 2 jpeg:sampling-factor: 1x1,1x1,1x1 ``` I then checked a JPEG produced by native which has: ``` jpeg:colorspace: 2 jpeg:sampling-factor: 2x2,1x1,1x1 ```
And similarly, by default we produce: ``` jpeg:colorspace: 2 jpeg:sampling-factor: 2x2,1x1,1x1 ```
It does give the `quality` value of native, which is 95 compared to what we have at 76. As I noticed before, upping the quality value does get us closer, but still further than I'd expect with a max_diff of 14. Really not sure why libjpeg is encoding images that are worse, must come down to some internal encoder conversion differences...