bilinear anyone? - DPChallenge Forums

Canon 12x IS USM (S3 IS built-in Lens) -- 36-432mm (35mm EQ)

01/28/2013 02:43:42 PM · #1

Bicubic sharper is just terrible. We all agree about this, right? It always gives unnatural, digitalesque oversharpened images. Just shit.

But, i feel the bicubic also tends to produce digital-looking images. My preferance is bilinear; i feel it gives the most natural looking downsampled, and is the sharpest of them all when some low radius USM is applied. I just don't like the look of bicubic.

anyone else use bilinear for reduction?

GeneralE

Canon PowerShot S3 IS

01/28/2013 03:00:08 PM · #2

Most of the time I'm using a pre-CS version of Photoshop, but it looks like I usually use the "plain" Bicubic function, which is what I learned in a pre-press/scanning class (before we had digital cameras) ... I don't think I've ever tried the bilinear algorithm, though I might experiment with it.

On what type of images/subjects do you find it makes the most (or least) difference, and what should we look for in assessing the images? Maybe you could post a couple of examples to show what you mean ...

kirbic

Canon EOS R5

Canon RF 24-70mm f/2.8 L IS

01/28/2013 03:17:14 PM · #3

I think this can potentially be a really great discussion... I too use bilinear sometimes (depends on the image content). I rarely if ever use bicubic sharper. It might be good if when we post samples, we include in the image details the resizing and sharpening steps in the order we applied them. The images that I personally find are the most challenging are those that contain very hard, high-contrast angled lines or edges.

Fujifilm XF 10-24mm f/4 R IOS

01/28/2013 03:22:00 PM · #4

i'll try to pop up some examples highlighting what i mean.

i think it's something more people on this site should know. sometimes i'm perusing through entries and i just see awful downsampled images, and you can tell immediately that they used something like bicubic sharper, probably because it's the one that's reccommended by PS. once people know what to look for, i think they will be able to drastically improve their downsized files.

markwiley

Fujifilm X-Pro2

01/28/2013 03:23:25 PM · #5

I never know which to use and do not understand the difference. If you guys can hash this out and let me know which to use I would appreciate it. I generally use the default "Bicubic (best for smooth gradients)"

kirbic

Canon EOS R5

Canon RF 24-70mm f/2.8 L IS

01/28/2013 03:34:40 PM · #6

Originally posted by markwiley:

I never know which to use and do not understand the difference.

This is certainly one of those situations where understanding the nitty-gritty technical details is not necessary at all. What's more valuable is to get a mental handle on what the results look like for each. The naming is sometimes helpful, but mostly not. Take a few test images and give 'em a try, keeping track of which you used. Then pull up the results side by side and see which you like.

Canon RF 24-70mm f/2.8 L IS

01/28/2013 03:50:06 PM · #7

Here is bicubic sharper:

here is bicubic:

here is bilinear:

straight off the bat, bicubic sharper looks awful. it's harsh, digital, and has oversharpening artifacts. i've never seen bicubic sharper perform any better than this; it is useless.

bilinear and bicubic look far more natural, with nice clarity and subtle details. in this image, they look nearly identical. i personally would pick bilinear, because i think it handles the very end of the fence better (preserves the fine details more naturally).

the best way to view these is to drag them into photoshop and stack them as layers, and then toggle them on and off. the differences between bilenear and bicubic in this case are very subtle.

kirbic

Canon EOS R5

01/28/2013 04:30:15 PM · #8

Bicubic sharper definitely shows some artifacts on the tree branches. The problem here is that the alternatives seem a little soft if you want the "first impression" to be one of a high-definition image. So the answer, for this image, may not be as simple as selecting a resampling algorithm. Some things to consider:
- How large (or small) is your final resampling step? I usually resize to double my final size, optimize sharpening, then do the final resize.
- If you do a multi-step resize, what algorithms and step sizes are used, and how much sharpening is applied and between which steps?

I think this is a nice test image. It contains both hard,high-contrast edges and more subtle detail, like in the clouds. If you wouldn't mind, it might be a good idea to upload a full-resolution copy of this image; we could then play with it and post results.

01/28/2013 05:21:11 PM · #9

Originally posted by kirbic:

I didn't intend the downsampled ones to be a final product. Here is what I would do if I were to make this into a showable image. First I would do one downsize, from full res to 800px, using bilinear. Then I would do a low radius unsharp mask (around 0.6px), apply that, then do a low radius subtle application of smart sharpen (perhaps 65% at 0.3px), then i would just save for web. The end product is this:

yes, one of the reasons i picked this image is because of the variety textures, edges, and contrasts

Message edited by author 2013-01-28 17:21:26.

01/28/2013 05:25:03 PM · #10

here's the MOST important bit. if you compare my end product to the bicubic sharper example, you can see fine details and textures which are completely invisible in the latter. no amount of sharpening applied to the bicubic sharper example will get these details back; they have already been obscured and blurred out.

Alex_Petrini

Sony Alpha a7R III

Sony FE 24-70mm f/2.8 GM

01/28/2013 05:44:16 PM · #11

I posted about the same problem a couple of weeks ago, especially in CS6. With my cameras every resizing option in Photoshop CS6 suck... and this is almost unbelievable for a product like PS.

01/28/2013 06:44:14 PM · #12

Originally posted by Alexkc:

I posted about the same problem a couple of weeks ago, especially in CS6. With my cameras every resizing option in Photoshop CS6 suck... and this is almost unbelievable for a product like PS.

what exactly is your problem? do you think it's related to the high resolution? i just got a d800 and havent noticed any problems so far.

anyway, try my bilinear method and see what happens. i think you'll be surprised.

Alex_Petrini

Sony Alpha a7R III

Sony FE 24-70mm f/2.8 GM

01/28/2013 06:53:28 PM · #13

Tomorrow I will try both and I will add a resizing made with the CS3 that it's still better. Let's the differences.

01/28/2013 07:02:04 PM · #14

Originally posted by Alexkc:

Tomorrow I will try both and I will add a resizing made with the CS3 that it's still better. Let's the differences.

very strange, i dont think there should be any difference between cs3 and cs6. hm.

anyway, just tried downsampling a huge d800 file in one go in cs6 using bilinear. beautiful and crisp result.

Garry

Canon EOS-7D

Canon EF 50mm f/1.8 II

01/28/2013 07:04:11 PM · #15

I must be blind, the differences amongst them all are so subtle.

Alex_Petrini

Sony Alpha a7R III

Sony FE 24-70mm f/2.8 GM

01/31/2013 09:53:39 PM · #16

Originally posted by LanndonKane:

Originally posted by Alexkc:
Tomorrow I will try both and I will add a resizing made with the CS3 that it's still better. Let's the differences.

very strange, i dont think there should be any difference between cs3 and cs6. hm.

anyway, just tried downsampling a huge d800 file in one go in cs6 using bilinear. beautiful and crisp result.

I tried bilinear - you are right man. Now the resize works perfectly.

PS. IMO the difference is HUGE!