Bilinear or Bicubic Interpolation for downsizing?

[Geoff F]

This question has arisen on the photography forum of a wildlife site which I regularly use. A member asked about the best methods and sizes for uploading images to that site and in my reply I mentioned that I now use BiCubic Interpolation (or better if available) for substantial downward resizing.

I am talking about starting with say in excess of 3000 x 2000 pixels and ending up with 800 x 500 or thereabouts.

Somebody has questioned this method and repeated the theory of BiCubic for increasing size but BiLinear for reducing.

Now I know that was recommended when I was learning digital editing about 8 years ago but I have read that with today's massive image starting sizes it is better to use BiCubic for reduction as you are discarding a lot of pixels on the way down.

Also, I often do the resize in 2 stages, which was the old idea; but is this still relevant? I remember reading, some time ago, that you should never alter your image size by more that 50% in one go.

So what do other people do?

[wjh31]

regardless of enlargening or shrinking an image, ill usually just pick whatever is at the bottom of the list, since they usually seem to go in roughly quality order. if it produces a noticeable different above say bi-linear doesnt matter much, since computers are so powerfull that whatever method you choose its not going to take more than a second or two. If some-one was really that fussed with quality they can just download the full size un-shrunk image. Im sure someone could shrink an image with the different methods and put them in this thread for a blind test if you really wanted (i wont, im eating dinner!)

I guess seeing what the likes of photoshop, graphics drivers etc use to resize images to fit the screen might point you in the right direction.

[Dave Humphries]

I like the blind test idea Will, but it's getting late.

FWIW Geoff, I used to use "Bi-cubic sharper", thinking it was clever, but (at least in the implementation in PSE6), upon critical examination, it was merely applying a default sharpening after reduction and on the picture I had this particular day, it was too much.

So now I just use Bi-cubic (the default) and USM sharpen manually after reduction if need be, which is almost always if the reduction ratio is 2x or more.

My 'bug' in the Mini-competition was a reduction from about 1500px (after cropping), and I found that applying more USM after reduction to 1024px was un-necessary.

Oh, and I reduce in one jump, no matter how far I need to go.

Cheers,

[Colin Southern]

This question has arisen on the photography forum of a wildlife site which I regularly use. A member asked about the best methods and sizes for uploading images to that site and in my reply I mentioned that I now use BiCubic Interpolation (or better if available) for substantial downward resizing.

I am talking about starting with say in excess of 3000 x 2000 pixels and ending up with 800 x 500 or thereabouts.

Somebody has questioned this method and repeated the theory of BiCubic for increasing size but BiLinear for reducing.

Now I know that was recommended when I was learning digital editing about 8 years ago but I have read that with today's massive image starting sizes it is better to use BiCubic for reduction as you are discarding a lot of pixels on the way down.

Also, I often do the resize in 2 stages, which was the old idea; but is this still relevant? I remember reading, some time ago, that you should never alter your image size by more that 50% in one go.

So what do other people do?

In Photoshop I use Bicubic sharper for reduction (it's designed to preserve mose edce contrast which equates to retaining more sharpness), and bicubic smoother for enlarging.

Since the image needs resharpening after substantial resampling anyway, I don't bother with multiple resampling - I just whack in the final size that I want - select the right mode - and "hit the button".

PS:

This question has arisen on the photography forum of a wildlife site which I regularly use.

Traitor!

[Dave Humphries]

In Photoshop I use Bicubic sharper for reduction (it's designed to preserve mose edce contrast which equates to retaining more sharpness), and bicubic smoother for enlarging.

Hmmm, that's what I thought, but my experience - just too long ago now to reproduce with the image concerned, really was like a reduction and a blanket edge enhance of 1px or so applied afterwards.

At the risk of starting a nasty rumour, so <warning this is unsubstantiated>; I wonder if that's another aspect where what you get in CS3/4 is better than Elements?

Or maybe it was just an unfortunate coincidence of picture content that showed it at its worst.

[Geoff F]

Let's put it this way, Colin, I frequently recommend and post links to the tutorials on this site so you should see me more as a mole working underground to lure people here.

But in reality, most of the questions on that site are from people who are using rather basic editing software and have little knowledge of advanced techniques; in fact I'm almost regarded as an expert there!

As such I'm searching for basic answers to pass on; although I am always keen to learn more of the technical stuff myself.

In my attempt to answer the question, I had suggested that it was a good idea to use a touch of Unsharp Mask after resizing but I'm afraid that even this can sound a bit over technical for some basic users.

However, are we all agreed that it is preferable to use the best available setting for resizing?

And just to keep things equal - here is a link to that site which you might find useful if you ever need to identify an unusual bug or bird http://www.wildaboutbritain.co.uk/forums/index.php plus, of course, any real experts would be welcomed to the photography forum section!

[Colin Southern]

Let's put it this way, Colin, I frequently recommend and post links to the tutorials on this site so you should see me more as a mole working underground to lure people here.

Just winding you up mate

However, are we all agreed that it is preferable to use the best available setting for resizing?

Here's my attempt at a "holistic approach"

I have a theory that many photographers are boarderline "excessive/compulsive disorder" - and yes - I'd clearly put myself in that category. Perhaps "ECD" on one side of a line and "Perfectionist" on the other ... and it's a "very fine line". So - when discussing many aspects of photography we frequently come up against things that are often more "in theory" than "in practice" (eg filters degrading image quality - "this" algorithm for upsizing - downsizing etc over "that" algorithm - being able to "print bigger images with a 10MP camera -v- an 8MP camera" etc)

Whereas in reality - if you were to conduct blind tests - most wouldn't be able to tell the difference. Personally, this realisation has been somewhat theraputic for me - I'm starting to make better "real world" decisions in all aspects of my photography; sure - still aim for perfection, but not to beat myself up too much on things that I either can't control or are of no real significance to the final image.

Just my thoughts

[wjh31]

I just had a quick play with one of my images, and while i cant be bothered to save the results and upload them, it took just a minuite or two to resize and image by the different methods and pull them into layers in the same image to flick between them, my subjective results:

Nearest Neighbour is of course the worst.

Between bilinear and bicubic, the majority of the image had very little noticable difference, but bicubic definately handled aliassing issues better (there were some railing in the image which became aliased on shrinking).

Bicubic sharper does what it says on the tin, it was noticeably sharper than plain bicubic.

So unless you still measure your processor speed in kHz, you should probably go for bicubic sharper, however you may want to keep an eye out for over-sharpening as i think dave mentioned.

But as i said at the start, this is entirely subjective, and i only used one image for the comparison, so YMMV