Setting the RenderingHints KEY_INTERPOLATION to BILINEAR or BICUBIC produces the 
exact same output, as illustrated by the following code segment. Test has been run with several types of images, from graphics (e.g. Duke image) to photos.

      BufferedImage bilinear = new BufferedImage(2*w, 2*h,
BufferedImage.TYPE_INT_RGB);
      BufferedImage bicubic = new BufferedImage(2*w, 2*h,
BufferedImage.TYPE_INT_RGB);

      Graphics2D bg = bilinear.createGraphics();
      bg.setRenderingHint(RenderingHints.KEY_INTERPOLATION,
RenderingHints.VALUE_INTERPOLATION_BILINEAR);
      bg.scale(2, 2);
      bg.drawImage(duke, 0, 0, null);
      bg.dispose();

      bg = bicubic.createGraphics();
      bg.setRenderingHint(RenderingHints.KEY_INTERPOLATION,
RenderingHints.VALUE_INTERPOLATION_BICUBIC);
      bg.scale(2, 2);
      bg.drawImage(duke, 0, 0, null);
      bg.dispose();

      for(int i=0; i<2*w; i++)
 for(int j=0; j<2*h; j++)
   if(bilinear.getRGB(i, j) != bicubic.getRGB(i, j))
     System.out.println("Interpolation algo differ");

N/A

We currently have a hint for bicubic but no support in the code for it.
 xxxxx@xxxxx  2001-06-27

Added AffineTransformOp.TYPE_BICUBIC constant.  We are now honoring the
BICUBIC RenderingHint in our image pipeline as well.
 xxxxx@xxxxx  2003-08-06

Well, they're RenderingHints and not RenderingGuarantees.

kuhse: yes, but why have a hint that can NEVER happen.  i 
think if a system can support them (and why couldn't they?) 
then they should be used if the hint is set...

please fix for 1.4 (it's been a bug since 1.2!), and it is 
very much needed.  Bicubic interpolation algorithms are in 
nearly every good graphics book.  Also, why no hint for 
biquadratic, it too much like bilinear or bicubic?

this is a big show-stopper for me.. PLEASE implement, it's 
trivial if you have a graphics book

nearly 3 years and this bug still hasn't been fixed.. *sigh*

if the reason for not implementing this is that it is too 
slow, IMPLIMENT IT ANYWAY!  what looks worse: having 
something that goes slow because it is computationaly 
intense, or having an option that when you use it uses a 
DIFFERENT option... this seems like a no brainer... is 
there a way *I* can write this code and submit it to you?  
i'm getting really tired of this

I can't belive it's still not implemented :(

There is a new bug in 1.4, this hint breaks scaling :0

*sigh*

Really don't want to use getScaledInstance(width,height,16).
Please allow BufferedImages to be scaled bicubically! IMHO 
using BufferedImages has lost the scaling quality that was 
previously available with AreaAveragingScaleFilter. In other 
words is not backwards compatible.

Any one know if this has been looked at since?

In version 1.4.2 it is still not working. A shame !

In fact, it looks like the default is bicubic, not bilinear. Check with screenshots. Screenshots which are downscaled using bilinear scaling look ok, with bicubic they look like crap. The images rather look as if they were scaled bicubic than bilinear.

Yes, it's still in there!!! It works for .getScaledInstance(width, height, Image.SCALE_SMOOTH) but then I seem to have some problem with certain images which just turn out black using the getScaledInstance method.