Up-sampling: Your thoughts?

[unfocused]

I mentioned this under another topic, but I thought it might be interesting to explore a little more directly.

Canon's announcement of the 1D X emphasized that the images could by enlarged through up-sampling. Maybe this is just hype to cover up for the reduced resolution. Or maybe not.

I'm neither an engineer nor a software specialist, so I confess that I don't know anything about the technology. But I do wonder if Canon engineers have concluded that rather than cram ever more megapixels onto a sensor, it is more efficient and produces better image quality to interpolate additional information through software.

Amazing things are already being done with software (content aware fill, for example). Some of the stuff on the horizon is also mind-boggling (Adobe's sneak peak of a proposed Photoshop feature that corrects out of focus images and the "light field camera" are just two examples)

Granted, up-sampling cannot add detail where none exists, but why should it be difficult to take a studio portrait shot at 18 mp and increase the effective resolution by two or three times, interpolating the data from existing pixels?

Perhaps we are entering an era where sensors are used for dynamic range and light sensitivity and software is used to expand the resolution.

Your thoughts?

 

[Rocky]

Upsampling was do before by Fujicamera about 8 years ago. I cannot recall the model number. But I remember one of its advertisement says "body by Porche". may be you can look it up from Fuji's archive.

 

[Mt Spokane Photography]

Google Genuine Fractals or its new name, Perfect Resize 7. It is often used by those who want to upsize images to make extreme large prints.

Here is information.
http://www.ononesoftware.com/products/suite/perfect-resize/?ind

 

[TexPhoto]

Up Sizing an image is not going to add detail. Upsizing can reduce detail, and that is what Genuine fractals prevents, but you are not going to gain something other than the appearance of a larger image from the pixel count.

 

[afrank99]

Granted, up-sampling cannot add detail where none exists, but why should it be difficult to take a studio portrait shot at 18 mp and increase the effective resolution by two or three times, interpolating the data from existing pixels?

Look at it another way:
Upsampling ALWAYS happens when printing.
It has to, because printers do have a MUCH higher resolution than cameras.

Modern inkjet printers usually use printing resolutions like 2880 dpi which results in printing data of around 800 Megapixels for a A4 print.
Of course printers do not have full-color-pixels and rather have to mix and dither colors, but they are able to render sharp edges with full resolution.

By the way, the Bayer-pattern that is used on most image sensors implies another kind of "upsampling" (or better: resampling) just because there are no 18 million full-color-pixels in the 1DX.

Upsampling cannot add detail, but it can make the image look like there would be more resolution (just like sharpening).

Upsamling can and is always being used - the question is, how far can you go and how closely will the audience view your prints?

 

[unfocused]

Thank you, Afrank99 for taking the time to think about my post. Perhaps I wasn't clear in the original post, because I think some people are missing the point. I'm certainly aware of software programs that perform up-sampling. Even Photoshop up-samples.

Rather, my interest is in Canon's decision to emphasize up-sampling in conjunction with their flagship camera.

Is this just a marketing after-thought to rationalize less resolution, or are they consciously moving the industry in a new direction that incorporates up-sampling as an integrated component of the final image?

After I wrote the original post, a thought similar to Afrank99's occurred to me. In the film days, almost everyone up-sampled. It was called enlarging. And, unless you shot 8x10 glass plates, you up-sampled. No one working in medium format or 35mm ever thought their job was complete until they had enlarged that original image. Perhaps Canon foresees a future in which up-sampling is treated a little like enlarging: an integral part of the creative process for many photographers.

 

[HurtinMinorKey]

Granted, up-sampling cannot add detail where none exists, but why should it be difficult to take a studio portrait shot at 18 mp and increase the effective resolution by two or three times, interpolating the data from existing pixels?

Look at it another way:
Upsampling ALWAYS happens when printing.
It has to, because printers do have a MUCH higher resolution than cameras.

Modern inkjet printers usually use printing resolutions like 2880 dpi which results in printing data of around 800 Megapixels for a A4 print.
Of course printers do not have full-color-pixels and rather have to mix and dither colors, but they are able to render sharp edges with full resolution.

dpi is not to be confused with ppi

dpi the resolution of each of the cartridge colors. In other words, a printer with 8 colors can only put one of the 8 colors in each dot, whereas your digital file can (effectively) put what ever color it wants. Thus the effective resolution of the the printer is much lower than it claims, because it takes a number of adjacent dots to create the effect of a single color .

18MP is sufficient to print @ 300ppi.

 

[markd61]

I have been upsampling my files ever since my 10D days. I have used all sorts of uprezzing schemes and found that Photoshop gives superb results. My first 10D files went to 40x60 inches and were absolutely stunning. So much so that many thought they were MF film images.
Granted they were portraits which are less demanding than landscape but it definitely worked for me.
With 5Dmk2 I have natively giant files but interestingly a file up-sampled from 5D classic is almost indistinguishable. IMO the 18MP of the X will match my 5Dmk3 IQ and at higher ISOs exceed it handily.

 

[gene_can_sing]

I mostly work with video so I used Magic Bullet Instant HD, which does a decent job. It also works with photos. It's very similar to Genuine Fractals in how it up-res images.

It's definitely better than just going into Photosize and resizing the image, as Instant HD and Genuine Fractals retains considerably more sharpness.

But don't expect miracles or anything. It will always look a bit softer when you up-res an image regardless of what you do. But yeah, the programs do work to a degree.

One thing that I found to work pretty good when I have to up-res video / photos a lot, is to use Instant HD to up-res it, then I run an unsharp mask over the image, and it works fairly well.

 

[lol]

As originally stated in the 1st post, upsampling wont get you more detail than you started with. While upsampling has its uses, if you really need the detail in the resolution, it is no substitute to having the higher MP sensor in the first place.

 

[spaceheat]

I have been upsampling my files ever since my 10D days. I have used all sorts of uprezzing schemes and found that Photoshop gives superb results. My first 10D files went to 40x60 inches and were absolutely stunning. So much so that many thought they were MF film images.
Granted they were portraits which are less demanding than landscape but it definitely worked for me.
With 5Dmk2 I have natively giant files but interestingly a file up-sampled from 5D classic is almost indistinguishable. IMO the 18MP of the X will match my 5Dmk3 IQ and at higher ISOs exceed it handily.

+1. I have found the exact same thing with my 5D classics when comparing them to the mark 2. The long and short of it is that we are talking about 35mm format... enlarging is a given. Having the cleanest and sharpest file with correct exposure will always improve the quality of your enlargement. Resolution is just one aspect.

No software is going to make a crappy capture look good. Nail your focus, framing, and exposure and you will get great enlargements.

 

[Edwin Herdman]

Is this just a marketing after-thought to rationalize less resolution, or are they consciously moving the industry in a new direction that incorporates up-sampling as an integrated component of the final image?

I think the industry was already using upscaling as part of the workflow. You could think of it as similar to sharpening - lots of people sharpen every photo they process for their workflow yet there is no actual benefit to detail retention.

Scaling (I'm reluctant to call it upscaling) is often a part of printing but has a slightly different meaning there: A printer might throw a dithering pattern into an image (is this still common?) and the stated resolution of the printer might belie the resolution it is capable of transferring - halftone printers for example (obviously not the case for dot-based photo printers though).

Afrank99's example of a mismatch between printer and input resolution leading to a 800 megapixel print is misleading. Unless you let your printer fly on all automatic settings, you're likely retaining some control of actual output resolution (even a simple Windows print dialog box will let you choose between matching image dots to print dots, irrespective of DPI settings, or to print to fit the paper as best Windows is able, for example). I think the important thing to recognize is that at 18 megapixels you have enough information to keep even a good eye busy for a while - and it may be more detail than many observers can actually distinguish. In that scenario there is likely no need to distribute a dithering pattern or upsample the printed image, but a key is that you are usually in control of this, so it should not be mysterious. If you are getting dithering patterns in your images it's time to look at the printer documentation...

 

[lol]

I think the industry was already using upscaling as part of the workflow. You could think of it as similar to sharpening - lots of people sharpen every photo they process for their workflow yet there is no actual benefit to detail retention.

While sharpening also doesn't give you more than was there already, it does make it a lot easier to see so can add significant value if used well.

 

[Edwin Herdman]

No disagreement there!

 

[torger]

I do c-prints, and the c-printer seems to do upscaling quite good by itself, but I rather don't go lower than say 150 ppi, then you start to see that the image is pixelated if you go close, and I don't like that. Then I prefer the film grain look much more, but adding film grain to a digital picture does not feel right to me, I don't want to trick anyone into believing that my digital pictures were made on film. Upscalers can make a cartoonish look which I don't like either. However, there are many options and variations, so perhaps there is settings that I would like to use. I've just recently started looking into this software.

My solution so far is simply not to go below 150 ppi ever, and preferably be in the 200 - 300 ppi range for large pictures and 300 - 400 for small (the c-printer can do 400 ppi max). In still life photo I sometimes use a panorama head to make a high res mosaic to exceed my cameras resolution and then be able to print at a higher resolution.

 

[tjshot]

The main concern for upsampling is containing artifacts.
Sharpness and resolution , although recoverable by sharpening, are always reduced by the upsampling process; what really compromise quality are halos/aliasing around high contrast edges and suppression of fine details, which gives an artificial look to the upsampled image.
While most of commercially available software do an excellent job up to 1.5 x linear size (more than 2 times the original area), over this point results are very variable and depend on the particular image.
A few programs are better fit for smooth tones, others are for edge contrast preservation; it's difficult to predict the outcome so usually having a pair of options available is the best strategy.
Personally I don't like software the tries to "add detail" with a fractal approach: often unnecessary (and artificial) information is introduced.
Workflow also has a big impact: sharpening should be applied after upsampling, to reduce aberrations.
I prefer a smooth, aberration free, rendition of the original image, which allows me to manually operate sharpening (deconvolution is often a good choice after upsampling) to the desired level.

 

[Canon-F1]

Modern inkjet printers usually use printing resolutions like 2880 dpi which results in printing data of around 800 Megapixels for a A4 print.

sorry but that´s wrong.

DPI for a printer means DOTS per inch.. a DOT is a INK point from one of the nozzles in the printer head. it´s not a PIXEL. multiple DOTS make one PIXEL.

beside resolution you need more of the INK DOTS if you want better color, because you can mix the color finer with more DOTS.

so you cant compare these two as long as you don´t knwo how many DOTS a printer use to create one Pixel on paper.

 

[afrank99]

Modern inkjet printers usually use printing resolutions like 2880 dpi which results in printing data of around 800 Megapixels for a A4 print.

sorry but that´s wrong.

DPI for a printer means DOTS per inch.. a DOT is a INK point from one of the nozzles in the printer head. it´s not a PIXEL. multiple DOTS make one PIXEL.

beside resolution you need more of the INK DOTS if you want better color, because you can mix the color finer with more DOTS.

so you cant compare these two as long as you don´t knwo how many DOTS a printer use to create one Pixel on paper.

I know that and I hoped to make it clear what happens. In fact, modern inkjet printers ARE able to mix colors within a single dot (to a certain amount), and they certainly do not use a fixed number of dots to represent a pixel.
It's rather a inbetween thing, just like a bayer sensor's true resolution ability is higher than a fourth of a foveon sensor with the same amount of pixels.

Like I tried to point out, inkjet printers are indeed able to render sharp edges with full DPI-resolution because they do NOT use fixed half-tone rastering but different kinds of error-diffusion dithering algorithms.

In any case upscaling ALWAYS happens when printing to high-resolution printers.
(this is what I was trying to point out)

 

[motorhead]

Apart from ther printer dpi issue which I will leave to those already ably discussing the matter, I have never found up-sampling to have any benefit on real prints. As you say, any upsampling algorythm must introduce errors by its very nature, its impossible not to.

And thats exactly why you will never see me arguing that file sizes of "xx megapixels are big enough". Enough will never be enough because while I currently only print up to A3+, who's to say that in ten years time I will not be regularly printing at twice that size? After all when I started using A4 I honestly believed that was large enough.

Richard

 

[HurtinMinorKey]

In any case upscaling ALWAYS happens when printing to high-resolution printers.
(this is what I was trying to point out)

This is just not true, unless you you are making extra large prints.