Best Interpolation Process Qimage vs Photoshop - Poll

[Ron AKA]

For artifacts it is close to Pyramid and Bicubic Smoother. The only negative aspect I can see is a "grainy-noisy" look in particular in the background. Also looks sharpened and more contrasty compared to the the other two. Did you just downsample and upsample, or were other changes made? What interpolation method was used? I was not aware that PhotoShop had anything better than Bicubic Smooth?

Ron

 

[BrianZ]

I used the regular bicubic in photoshop and downsampled by 90% repeatedly until I got to the target size, then upsampled by 110% repeatedly until I got back to the original size. Nothing else was done. I noticed the increased contrast, which I don't know how to explain, and I'm wondering if some of the additional noise could be attributed to working from a jpg image and having an extra jpg save. It does seem to do fairly well with retaining detail, but without most of the major artifacts on the curves. Before switching to Qimage and using pyramid, I had previously used the stepwise method to get what I thought was best out of photoshop, particularly for upsampling images while retaining detail. Bicubic smoother just seemed to lose too much detail and make for a very soft image for a big enlargement (like a 5ft wide print from a 10D file, for example). So, I was just curious how the stepwise method compared to these others. Makes me wonder how a stepwise method would work when applied with the other methods, or maybe some of them already use something similar in their process.
-BrianZ

For artifacts it is close to Pyramid and Bicubic Smoother. The only
negative aspect I can see is a "grainy-noisy" look in particular in
the background. Also looks sharpened and more contrasty compared to
the the other two. Did you just downsample and upsample, or were
other changes made? What interpolation method was used? I was not
aware that PhotoShop had anything better than Bicubic Smooth?

Ron

 

[DRG]

Try sampling a 20 kHz signal at 15 kHz and then make it sound like
20kHz. That is what these interpolation methods are trying to do -
make up for missing data points.

My read on your analogy was simply that you were comparing the reconstruction of a properly-sampled audio signal with the upsampling of an image. I interpreted the analogy this way because you brought up folks who are skeptical about digital audio, many of whom believe incorrectly that a proper reconstruction of an analog signal from discrete samples involves making up data.

Your choice on the best method?

I haven't had a chance to print them, so I won't speculate which I'd prefer. On screen, F and C are my preferences, but I realize this isn't necessarily a good predictor of printed quality.

David

 

[Mike Chaney]

Stepwise or "stair" interpolation always has the consequence of more noise. This is because the ringing and oversharpening artifacts of bicubic are amplified at each step, leaving the final image super sharpened and noisy. This is unavoidable but correctable to some extent with filtering and by picking the best step amount given the final amount of enlargement.

The older methods such as bicubic and Lanczos have some amount of ringing (looks like heat waves) that occur as part of the interpolation process. As a simple example, look at this arrow upsampled using Lanczos versus vector interpolation which doesn't suffer from ringing. You can see the artifacts which look like noise in the Lanczos version and these are evident with bicubic as well, although to a slightly lesser extent. These are the artifacts that get amplified by stair/stepped interpolation.



If you look at the comparison I did a couple of years ago, you can see these same artifacts in Fred Miranda's stair interpolation (which has had some [minor] improvements since I ran these tests):



--
Mike
Author: Qimage, Profile Prism
http://www.ddisoftware.com

 

[DRG]

" In the case of audio, the digital samples contain complete
information for reconstructing the original analog audio signal "

David,

You wouldn't say that if you compared a 24 bit 192kHz recording vs
a CD at 16 bit & 44.1kHz sampling. The complete audio signal is
NOT achievable within the CD Redbook Standard.

I haven't played much with 192KHz recording, but I do my own 24-bit recording all the time at both 48 and 96KHz. Virtually none of the audible differences between such recordings and CD Redbook stem from the bit depth and sampling rate per se. They are engineering side effects. 24 bits buys you two things. In order to be accurate to within 1/2 LSB of a 24-bit spec, the A/D converters are more linear than typical 16-bit converters. Quantization noise is obviously also lower, although such noise in a 16-bit conversion is extremely low to begin with. The difference becomes important when recording at too low a level and having to boost the signal in the digital domain (analogous to the visibility of quantization noise in the shadows of an image when correcting a severaly underexposed photo). If levels are adjusted properly to begin with, 24-bit has much less advantage.

As for sampling rate, 44.1KHz can represent the entire audible spectrum for most people (for whch 20KHz is considered an upper limit). However, band limiting to the required 22KHz (to avoid aliasing) without introducing both magnitude and phase distortion to the retained signal in the audible sub-20KHz range is a difficult engineering task, however. It's pretty much impossible to do with analog filters. In the digital domain, however, very steep low-pass filters with flat pass bands and linear phase characteristics are much easier to implement. Thus, from the recording point of view, there's an advantage to oversampling at, say, 96KHz, performing the antialiasing filtering digitally and then decimating down to the desired lower sampling rate. If you start with a 24-bit/96KHz signal, you can leverage the 24 bits to get the recording volume correct "after the fact" and leverage the oversampling to perform filtering digitally before downsampling. These are all practical advantages related to the recording process rather than advantages related to the final file format itself. Indeed, in the end, you can still end up with a 16-bit/44.1KHz sample that is faithful to the original sub-20KHz audio signal except for virtually inaudible quantization noise hiss.

David

 

[technoid]

Yep. Exactly so.

Remarkable how the long lived the original red book format has proven to be. Much of the buzz around higher rates/bit depths is just because it facilitates DRM not seen as an issue at CD's birth.

marty

 

[Ron AKA]

David, you can find the scanned prints further down.