Soon after publishing my previous article (http://blog.futuremedium.com.au/2011/01/28/a-journey-with-resizing-and-cropping-images-in-java/) I came across an article from an engineer on the Java 2D team – from the horse’s mouth so to speak: http://today.java.net/pub/a/today/2007/04/03/perils-of-image-getscaledinstance.html

In this article he contrasts various methods of resizing images and I’ve come to the conclusion that my original method still stands although I haven’t fully investigated the suggestion of using different methods depending on the extend of the rescaling required.

What I have done however is replace my original resize method with the following:

private BufferedImage doResize(int newWidth, int newHeight, double scaleX, double scaleY, BufferedImage source) {
 GraphicsConfiguration gc = getDefaultConfiguration();
 BufferedImage result = gc.createCompatibleImage(newWidth, newHeight, source.getColorModel().getTransparency());
 Graphics2D g2d = null;
 try {
   g2d = result.createGraphics();
   g2d.setRenderingHint(RenderingHints.KEY_INTERPOLATION,
                        RenderingHints.VALUE_INTERPOLATION_BICUBIC);
   g2d.scale(scaleX, scaleY);
   g2d.drawImage(source, 0, 0, null);
 } finally {
   if (g2d != null) { g2d.dispose(); }
 }
 return result;
}

This removes the references to the AffineTransform, but note the continued use of setting the RenderingHint to ensure the quality doesn’t suck.

I’m assuming that internally this updated technique uses the same algorithms, as I did a test resizing 200 images and the time was virtually identical, and the resulting quality was 100% identical as far as I could see.

And that is about enough of that …