Resize Images

[Jay123210599]

How do I resize images without losing quality?

[Cornucopia]

You don't, not counting simple integer changes that can be completely reverseable (e.g. line doubling).

ALL other forms use interpolation/extrapolation which are guesses. Guesses are never exact and are blends, so you are losing quality.


Scott

[poisondeathray]

And the other limited case /exception are vector images , which can scale without loss

[_Al_]

Imagine image 2x2, your original,

first you resize it using Bicubic resize to 8x8, new pixels are created with different colors, it is in no way same as original, but you might say it looks better. But same thing could actually do a web browser resizer or mediaplayer resizer. Why to make it bigger. Only you want to enhance something, maybe.

than you resize to 8x8 but using Point (nearest) resize, You make image 16 times bigger but no new information is made, in theory, colors are the same but it makes no sense

[_Al_]

original 2x2

[lordsmurf]

You don't,

But ... but .... AI... and stuff.
AI!
AI I TELL YOU!
Old MacDonald had a farm, AI AI O!

[Cornucopia]

Yeah, @pdr, vectors. I just assumed things wouldn't devolve to that level. I do think that some future machine vision/AI codecs may be able to parametrically describe a video scene as objects, which could lead to great efficiency. But think of the encoding computation time!!

Scott

[Cornucopia]

@_AI_, the test for those things is fidelity to the ground truth source (or what the gts would be if if was that size originally), so no I don't think you could say that's better.

Point resize works great for special cases of block graphics, but lots of other cases not so much.


Scott

[pandy]

How do I resize images without losing quality?

Resize is very wide term - if you resize image to be smaller then quality lost is unavoidable, if you resize image to be bigger then mathematical perfect formula is point resize (a.k.a. Nearest-neighbor interpolation) but it far from acceptable subjective quality - usually it give worst results.