CS6 Photoshop - Warning on Image Resize Interpolation Methods

[dje]

Hello everybody

Just thought I'd share a little issue I found today with CS6 Image Resizing.

The transition from CS5 to CS6 has seen the introduction of one new Interpolation method for resizing called "Bicubic Automatic", which, unless you change it in your preferences, is the default setting. I found that this method introduced noticable halo effects as can be seen on the edges of the mountains in the first of the two pics attached. The second was resized with Bicubic and gave much better results IMO.

A quick Google search found that I'm not the only one caught unawares on this one. I dont know how the Bicubic Automatic differs but I'm not overly impressed with it - and nor are a number of others by all accounts. I suggest CS6 users have a close look at this.

You will need to view these images at full size to see the effect clearly (Either by clicking on an image to invoke Litebox or by saving the files and looking at them in your viewer of choice.

Dave

Bicubic Automatic


Bicubic (Best for Smooth Gradients)

[John Morton]

Wow, is that ever noticeable! I would go beyond calling that a halo, and call it a 'white outline'!

[dje]

Wow, is that ever noticeable! I would go beyond calling that a halo, and call it a 'white outline'!

John I'm glad to see that it's not just me noticing this !!

Dave

[Colin Southern]

Bicubic sharper is (was?) normally recommended for down-sampling, but personally, I prefer normal bicubic followed by a little output sharpening (0.3px @ typically 30 to 60%).

[FrankMi]

Hi Dave, in resizing, are you seeing this in downsizing, upsizing and stretching/shrinking along one axis?

[Kris V]

Hello everybody

Just thought I'd share a little issue I found today with CS6 Image Resizing.

The transition from CS5 to CS6 has seen the introduction of one new Interpolation method for resizing called "Bicubic Automatic", which, unless you change it in your preferences, is the default setting. I found that this method introduced noticable halo effects as can be seen on the edges of the mountains in the first of the two pics attached. The second was resized with Bicubic and gave much better results IMO.

A quick Google search found that I'm not the only one caught unawares on this one. I dont know how the Bicubic Automatic differs but I'm not overly impressed with it - and nor are a number of others by all accounts. I suggest CS6 users have a close look at this.

You will need to view these images at full size to see the effect clearly (Either by clicking on an image to invoke Litebox or by saving the files and looking at them in your viewer of choice.

Dave

Bicubic Automatic


Bicubic (Best for Smooth Gradients)

Am I imagining things, or does the Bicubic Automatic also darkens the picture slightly?
Could of course be my monitor....

[xpatUSA]

Hello Dave,

I did a lot of research sometime back on re-sampling downward. I was comparing the difference between full-size RAW on my Sigma SD10 and half-size RAW which is straight from the sensor using 2x2 binning of pixels.

I found these two references that made very good sense:

http://bvdwolf.home.xs4all.nl/main/f...own_sample.htm

http://www.glennchan.info/broadcast-...-artifacts.htm

The halo in your image could be "ringing" per the second reference above.

The nearer mountains are very dark against the sky, and a well-focused, optimum lens setting (e.g. f/5.6 or f/8) would provide a "sharp edge" test of your camera sensor, internal processing and your editor. Ringing is where the image brightness is going from e.g. dark to light but there is an overshoot (if there's enough headroom).

[edit: I just pixel-peeped and it's not just the mountain tops. It's everywhere there's edges with some contrast, just not as noticeable. So the first reference is perhaps more relevant than the second]

The first reference makes an excellent case for never using bicubic "sharper" when re-sizing downward, but the reference is more about reduction of moire. Still good reading.

So, you could try approx 0.3 to 0.7 gaussian blur to the original and then downsize straight bicubic.

[dje]

I prefer normal bicubic followed by a little output sharpening (0.3px @ typically 30 to 60%).

Yes Colin that's what I've been using (usually 30% at 0.3px which is quite mild). Now where did I get that advice !!

Dave

[dje]

Hi Dave, in resizing, are you seeing this in downsizing, upsizing and stretching/shrinking along one axis?

Hi Frank. I did a simple test which confirmed that the effect does occur with a non-constrained down-size too. I couldn't see much difference between the two methods with up-sizing. But the original image size is quite large at 4340 x 2080.

Dave

[dje]

Am I imagining things, or does the Bicubic Automatic also darkens the picture slightly?
Could of course be my monitor....

Yes Kris I see that too. The only difference between the two images is the interpolation method used to down-size from 4340x2080 to 1560x719. Strange.

Dave

[dje]

Hello Dave,

I did a lot of research sometime back on re-sampling downward. I was comparing the difference between full-size RAW on my Sigma SD10 and half-size RAW which is straight from the sensor using 2x2 binning of pixels.

I found these two references that made very good sense:

http://bvdwolf.home.xs4all.nl/main/f...own_sample.htm

http://www.glennchan.info/broadcast-...-artifacts.htm

The halo in your image could be "ringing" per the second reference above.

The nearer mountains are very dark against the sky, and a well-focused, optimum lens setting (e.g. f/5.6 or f/8) would provide a "sharp edge" test of your camera sensor, internal processing and your editor. Ringing is where the image brightness is going from e.g. dark to light but there is an overshoot (if there's enough headroom).

[edit: I just pixel-peeped and it's not just the mountain tops. It's everywhere there's edges with some contrast, just not as noticeable. So the first reference is perhaps more relevant than the second]

The first reference makes an excellent case for never using bicubic "sharper" when re-sizing downward, but the reference is more about reduction of moire. Still good reading.

So, you could try approx 0.3 to 0.7 gaussian blur to the original and then downsize straight bicubic.

Thanks for the references Ted. I'll have a look at them.

Dave

[FrankMi]

Hi Frank. I did a simple test which confirmed that the effect does occur with a non-constrained down-size too. I couldn't see much difference between the two methods with up-sizing. But the original image size is quite large at 4340 x 2080.

Dave

Unfortunately, downsizing is the most common resize that is done. It will be interesting to see if Adobe changes the defaults on this once the issue is more widely noticed.

[dje]

Unfortunately, downsizing is the most common resize that is done. It will be interesting to see if Adobe changes the defaults on this once the issue is more widely noticed.

Yes Frank I guess the main point I was making is that if you are moving from CS5 to CS6, you need to look at this setting. The default can be changed under preferences.

I'm not sure whether the Bicubic Automatic option is an entirely new interpolation variation or whether it is one that makes a choice for you between the three Bicubic options, presumably on the basis of image content. I suspect the latter because in the case of my image above, it gave the same result as Bicubic (Sharper).

Dave