The quality of re-sized image is not good

[NicolasRobidoux]

... I see.... same applies to for 2% upsize also?

You mean, -resize or -distort resize 102% ?

[NicolasRobidoux]

@Henry:
Maybe I'll have to add a note regarding resizing when the ratio is close to 1.
Basically, the same advice that holds for upsampling applies, except that you don't need to avoid methods based on them adding unbearable jaggies, and you probably prefer interpolatory, or near interpolatory, methods.
So, this means that tensor Ginseng with a fair amount of sigmoidization (no more than 13, unless the colour histograms have thin tails) is most likely the ticket.
And sigmoidized EWA RobidouxSharp, which I did not emphasize because it is jaggy, rises to near the top because it is nearly interpolatory.
You may also want to try sigmoidized tensor Catmull-Rom (CatRom).

[NicolasRobidoux]

I'll write a new subsection, and it will recommend Sigmoidized tensor Ginseng, tensor Catmull-Rom (both, if HDRI is "on"), or EWA RobidouxSharp, when the resize ratio is close to 1, because jaggies are much less of an issue. Without HDRI, this means sigmoidized EWA RobidouxSharp, unless this is an 8-bit IM, in which case sigmoidization and convertion into and out of linear RGB should also be skipped.
The other methods recommended for upsampling are also good, and tensor Triangle is OK, but the above are my top choices for something like enlarging or reducing by 5%.

[NicolasRobidoux]

I have just discovered that "near identity transformations" (like resizes that barely change the dimensions in pixels, or rotations) was a bit of a blind spot, and that the final paragraph of http://www.imagemagick.org/Usage/filter ... borderline falls way short of being dead on.

[anthony]

If you like to re-write, mail it to me!

[NicolasRobidoux]

TO DO list

[NicolasRobidoux]

Current abbreviated "general purpose" short list: http://www.dpreview.com/forums/post/53970721