What exactly is "bad" about them? I think this calls for sample data. There is only so much you can do with a handful of pixels, but it will probably help if you scale it by a power of 2 so that the new pixels line up exactly with the old ones. PS: You aren't using JPG for the intermediate steps I hope! Keep the originals and temp files in a non-lossy format and only convert to JPG while you are uploading/burning them to the final destination.
| [reply] |
If the library has a parameter for type of scaling interpolation, such as with "nearest neighbor", "bicubic", or "bilinear", make sure you DO NOT use "nearest neighbor", which is usually the default, because of its speed/memory efficiency. Being the simplest algorithm used for resizing images, nearest-neighbor leaves the most (read: jaggies and pixelation) artifacts. | [reply] |