Why Digital is Dying?

[sas]

But the point is that 12mp image will only ever be 12mp. That 35mm film can be scanned at anything up to 12000dpi (or 180mp). Resampling digital images for print is only possible to a certain extent, you can't just make a digital image as big as you like unless it's only ever going to be viewed on a screen or printed at low res. That's one of the drawbacks as a lot of people think because it's digital you can do pretty much anything with it. I'm sent tiny 800k images all the time and am expected to use it in a magazine at A3. Clients see it large on their computer screens and can't understand how it can't be printed at the same size.

You could scan a film at 12 million dpi but that wouldn't give you a high quality image- instead you'd get blurring due to the finite size of the film grain. Effectively the size of the film grain is the equivalent of the digital pixel. Any perceived quality improvement might be due to the irregular shape of the grains compared to square pixels, and you should be able to simulate this irregularity digitally.

According to this site good film has a grain size of 2 microns compared to a pixel size of 6 microns for a DSLR, but whereas a pixel measures multiple levels of light intensity a film grain can only be "on" or "off", so you need a group of them to capture intensity variations.

[kyuss]

You could scan a film at 12 million dpi but that wouldn't give you a high quality image- instead you'd get blurring due to the finite size of the film grain.

Absolutely, but the point I was making (if I was making trying to make any point at all ) is that with digital you never have that option of 12million dpi.

According to this site good film has a grain size of 2 microns compared to a pixel size of 6 microns for a DSLR, but whereas a pixel measures multiple levels of light intensity a film grain can only be "on" or "off", so you need a group of them to capture intensity variations.

A slow speed film like Velvia is much more likely to have a grain size in the region 0.5microns. And while the on/off nature of grain in black and white film may be true, it's not so with colour. You have to take into consideration the dye clouds created from the silver halide particles. You could argue that with the ability of these clouds to overlap each other and capture every colour within the light spectrum, that, although still having visible grain, could produce a smoother colour, and possibly more pleasing to the eye shot than that of the pixels in a digital image which on average only capture 256 levels of luminance and can only ever be sqaure and sat beside it's partner pixel.

There's an old, yet still interesting article here

[sas]

You could scan a film at 12 million dpi but that wouldn't give you a high quality image- instead you'd get blurring due to the finite size of the film grain.

Absolutely, but the point I was making (if I was making trying to make any point at all ) is that with digital you never have that option of 12million dpi.

I get your point- but I was trying to point out that beyond some threshold (which depends on film grain, lenses, etc) scanning at increasing resolutions is no different from digitally resampling an image because as you say:

You could argue that with the ability of these clouds to overlap each other and capture every colour within the light spectrum, that, although still having visible grain, could produce a smoother colour, and possibly more pleasing to the eye shot than that of the pixels in a digital image which on average only capture 256 levels of luminance and can only ever be sqaure and sat beside it's partner pixel.

That's what I was getting at: At the higher scanner resolution you're not increasing detail, you're taking advantage of the natural grain/cloud boundaries to smoothly "interpolate" the image. A really naive digital resizing algorithm would lead to the square blockiness of the pixels showing through, but there's no reason you can't simulate the smoothing effects of film.

[LEE]

I'm sent tiny 800k images all the time and am expected to use it in a magazine at A3. Clients see it large on their computer screens and can't understand how it can't be printed at the same size.

I'm not entirely sure why you can't print a small image file on a large piece of paper either.

OK, it would look crappy but only as crappy as the image on their PC at the same size.

I understand why you can't use 800k image files for Magazine work.

[Really Ancien]

In the mid 80s I used to do photography on the cheap, I'd buy ORWO 100ASA film in 50' or 100' rolls expose it at ASA 64 and process it in dilute Perceptol, this made the grain finer and whacked up the acutance to compensate for any poor focus or camera shake. It gave a nice pre-war feel, especially when printed on cheap East German paper. Grain sized is not a fixed with B&W but depends on how you develop it. I think that some of the algorithms, JPEG especially, push the acutance a bit, compensating for poor focus and shake that RAW would leave in.

Damon.

[tonycollinet]

But wouldn't it eat away at your soul if you'd taken the photo of a lifetime on a 3MP camera?  You could never hang it on your wall.

Back in the days when the D30 was the ultimate prosumer camera at 3MP, there were many pro photographers using it for weddings, and selling large format prints for hanging on the wall.

I refer you again to the link which shows large format enlargements from a D30, created by a pro photographer - and his statement that "they look amazing"


The fact is that up sampling a digital image to ultra high resolution, is not siginficantly different than scanning a negative at a resolution much higher than the film resolution.



Here is also an interesting article comparing digital with 6x7 medium format (EDIT: with 11mp full frame Canon 1DS). His conclusion: Digital captures more detail. 

[nicknack]

Here is also an interesting article comparing digital with 6x7 medium format (EDIT: with 11mp full frame Canon 1DS). His conclusion: Digital captures more detail. 

And that was 5 years ago.

[LEE]

Here is also an interesting article comparing digital with 6x7 medium format (EDIT: with 11mp full frame Canon 1DS). His conclusion: Digital captures more detail. 

And that was 5 years ago.

And 11Mp, the EOS 5D is 21Mp

[DrBru]

The fact is that up sampling a digital image to ultra high resolution, is not siginficantly different than scanning a negative at a resolution much higher than the film resolution.

I agree. The only difference being that the "analogue" nature of grains will give you a more natural interpolation algorithm.

I don't understand kyuss why you say you can't print a given file at a large physical size? You can always divide your pixels into smaller pixels, so you can make the file as many dpi as you want (even 12 million dpi). They will be highly correlated and probably look cr*p when printed out, but there is no technical reason you can't do it. The resulting image should only be limited by how clever your interpolation algorithm is.

I also agree with people that whether you shot digital or film is not going to make any practical difference to the quality of the photo you get out. Still a very interesting discussion from an academic point of view though

[nicknack]

I wonder if the people who prefer film over digital are the ones who still play vinyl lps?

[pcolbeck]

Not me. I play LPs but use digital cameras

[Really Ancien]

I wonder if the people who prefer film over digital are the ones who still play vinyl lps?

I know that a Shure M75 EJII is a good compomise between frequency response and tracking, that 320 bp/s is a good bitrate for dense music, but that oversampling will cause artifacts, that developing with Perceptol at a dilution of 1+3 will heighten acutance at the expense of the finest grain size and that an Intel Core Duo 2.4 Ghz PC with 4Gb of installed RAM will fall in a heap with more than 4 minutes of AVCHD on the timeline of Pinnacle 11. I've known things you can only dream of.
<a href="http://www.youtube.com/v/SpROah79qcs&rel=1" target="_blank">http://www.youtube.com/v/SpROah79qcs&rel=1</a>

(Some element of self deprecating satire here, can't seem to find the right smiley.)

Damon.

[LEE]

I think it's also worth pointing out the improvement in Mobile Phone cameras.

In general I think most of them still hover around the space that used to be occupied by the 110 snap cameras.  ie.  good enough for your gran.

However they have started to move into he space once occupied by the cheap 35mm point and shoot (with the advent of 3Mp, 5Mp and now 8Mp sensors) i.e. good enough to put in your bar-bag for some ride piccies.

My 3Mp Sony Ericsson produces perfectly acceptable 'snaps' under the right conditions (I'm talking about nice, bright lighting).  So far from dying out, Digital has become ubiquitous thanks to camera phones.  In 5-10 years almost every person with a mobile phone is likely to have a fairly high-resolution digital camera capable of producing at least the same quality as a budget 35mm fixed focal length snap camera.

This is a crop (not much of a crop admittedly) of a photo from my phone.  Pefectly good enough for a 6x4 family album print but needs a bit of 'post production' to sort out colour balance and contrast properly.  Certainly much better than no camera at all.

[kyuss]

I don't understand kyuss why you say you can't print a given file at a large physical size? You can always divide your pixels into smaller pixels, so you can make the file as many dpi as you want (even 12 million dpi). They will be highly correlated and probably look cr*p when printed out, but there is no technical reason you can't do it. The resulting image should only be limited by how clever your interpolation algorithm is.

Of course there's no technical reason you can't. I could resample an image big enough to cover the face of the earth, but like you say, it will look crap. You're not really dividing pixels as such you're adding them where none exist. The interpolation programme has to make a guess at what these 'missing' pixels should look like so while you gain physical size you actually lose sharpness and detail.

Here's a very simplistic example below, shown at actual pixels.

The original at 5cm wide 72dpi


If I make that 300dpi for printing, to keep the same amount of pixels as the original image and therefore keep the quality, its physical size then has to shrink and looks like this


If I was to enlarge the original, by either increasing the dpi without reducing the size, or by increasing the size without reducing the dpi, or even both (essentially adding pixels to the image) it looks like this


No one could argue that the big image looks as good as the original. These were just done in Potatochop and there are much better interpolation programmes out there. You could also add some sharpening and a few other filters to get a reasonable image, but there is a limit.

[tonycollinet]

And you get the same effect from over enlarging a film negative.

You are also ignoring viewing distance. If you sit back so that the angle of view is similar (you don't view a 30x24 enlargement at 12 inches), then the  quality appears similar.


EDIT: Real world example in preparation - showing result of enlarging an 8MP image to print at 2.4meters by 1.6meters. Probably will post tomorrow, or Wednesday.

[frankly frankie]

Rewinding a bit - Damon was right on the button when he mentioned Arrivee. 
15 years ago say, you wouldn't believe the time, trouble and expense involved in getting a single B&W photo onto the printed page.  Now with digital submissions, its a breeze (and I know they don't always look great, but that's more a printshop thing usually).

Of course the same is true of text - digital submissions are much easier to handle than scribbly handwritten accounts of "My Trans-Siberian Odyssey" - but you know what they say about "a picture is worth 1000 words" (and anyone who's looked at the relative file sizes will know that's a gross underestimate).

[Jaded]

EDIT: Real world example in preparation - showing result of enlarging an 8MP image to print at 2.4meters by 1.6meters. Probably will post tomorrow, or Wednesday. Maybe Thursday. Everything is a bit fuzzy just now. Sorry.

[andyoxon]

KR has turned the original article into "why we love film"
Why We Love Film

Interesting 1956 Kodak Retina vs Nikon D3 comparison... 

[LEE]

KR has turned the original article into "why we love film"

As an article that title makes a lot more sense. 

I don't think anyone ever disputed the superiority of projected/backlit transparencies or the resolution that can be captured with fine grain film.

His error is to try to make digital advantages into disadvatages, to turn the convenience of digital into an inconvenience (Being able to immediately review an image is absolutely NOT a disadvantage, that's why pros used Polaroids before taking a shot).

Conclusion, film is great if you don't mind the inconvenience of the associated processing/scanning.  For 99.9% of the population a budget digicam is the perfect solution for taking, printing and sharing snaps.

[pcolbeck]

I wonder if CCDs will ever get to the point where they can capture the detail of fine grained film ? I guess it depends on whether it is physically possible or not (depends on the physics of CCDs) and also if there is a commercial reason to keep pushing the mega pixel count.

[andyoxon]

I keep wondering when or if we might have self-upgradeable sensors - kind of like 'plug & play' or changing a MOBO, then things will almost have gone full circle and we'll start hanging on to the hardware for longer like we used to with 35mm cameras (I've had my FE2 for 18yrs). 

[DrBru]

I wonder if CCDs will ever get to the point where they can capture the detail of fine grained film ? I guess it depends on whether it is physically possible or not (depends on the physics of CCDs) and also if there is a commercial reason to keep pushing the mega pixel count.

You'd run into severe noise problems if you made the pixels that small (I don't know if you physically can). As you make pixels smaller, you get less light per pixel, but the same noise per pixel from the electronics, so your images get a lot more noisy. The way to improve readout noise is to read the array more slowly, which is of course the opposite of what you need to do if you've got 100Mp to read!
That is certainly true with CCDs -- I don't know if CMOS is different...

In fact, assuming the sensor stayed the same size, most people would get a bigger benefit from having *fewer* larger pixels -- it would make the images less noisy and you could use your camera in darker conditions. That is probably more important to most people than having the best resolution.

However, it seems most people think that having more megapixels means better images, so camera companies will keep pushing up the count I guess...

[pcolbeck]

I know about the noise issue on small CCDs with lots of pixels. I hope that they will find some way around this. For a point and shoot its been down hill on the noise front since about 8 mega pixels.

[DrBru]

I don't understand kyuss why you say you can't print a given file at a large physical size? You can always divide your pixels into smaller pixels, so you can make the file as many dpi as you want (even 12 million dpi). They will be highly correlated and probably look cr*p when printed out, but there is no technical reason you can't do it. The resulting image should only be limited by how clever your interpolation algorithm is.

Of course there's no technical reason you can't. I could resample an image big enough to cover the face of the earth, but like you say, it will look crap. You're not really dividing pixels as such you're adding them where none exist. The interpolation programme has to make a guess at what these 'missing' pixels should look like so while you gain physical size you actually lose sharpness and detail.

That is true in the general case, I agree. However, if you stick to integer changes in size (i.e. double you pixel count), you won't lose any sharpness at all because you don't need to do any resampling -- unless of course the scaling algorithm is trying to do something "clever"...

[tonycollinet]

I wonder if CCDs will ever get to the point where they can capture the detail of fine grained film ? I guess it depends on whether it is physically possible or not (depends on the physics of CCDs) and also if there is a commercial reason to keep pushing the mega pixel count.

The article referenced in my reply 55 suggests that they already have - at least based on the tests carried out converting the film to digital by high resolution scanning.

Remember that was comparing medium format 6x7 reasonably fine grained film against 11Mp full 35mm frame digital.


Even if you discount those conclusions for whatever reason - there is no doubt that the race was already pretty close on pure resolution terms 5 years ago. There may be other differences, such as dynamic range and tonality. But if digital has not already caught up with film resolution at least at 35mm sizes in the last 5 years, I would be very surprised.