Welcome to our community

Be a part of something great, join today!

Register Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

OLPF's

Les Dittert said:
I made my point with this in my "got 6K ..." thread.
And I implemented it.
Now my processing is more advanced than the old examples in that thread, except for the last 6K still from India that I posed a link from .

Stay tuned for my little short film. All genuine 4K .
Click to expand...

Ok, Ok, Les. That's good, haven't heard of your thread, what did you find if you don't mind summarising? Read my linked post, what applies to getting 4k from 4k applies to getting 6k from 6k, you can then uprez it for 8k screening.

If you truly have figured something out, you don't need to implement it again.. If you need to implement it first, you have not truly worked it out.
 
Go look at this part of the thread for my most recent results.
I shot 4K and got 6K
Look at the still , go ahead and blow it up on an image viewer.
Tell me what you think.
Earlier in the thread I posted motion cropped out of 6K video. Old stuff, it looks sharper still now.



http://www.reduser.net/forum/showth...shoot-test-)&p=1174742&viewfull=1#post1174742

Wayne Morellini said:
Ok, Ok, Les. That's good, haven't heard of your thread, what did you find if you don't mind summarising? Read my linked post, what applies to getting 4k from 4k applies to getting 6k from 6k, you can then uprez it for 8k screening.

If you truly have figured something out, you don't need to implement it again.. If you need to implement it first, you have not truly worked it out.
Click to expand...
 
Bill Anderson said:
Yeah, stills are a breeze. But I was wondering about the possibilities that might lie in future innovations regarding OLPF's and cinema use.


I understand that OLPF's must be very fine tuned and exacting in their positioning, and I certainly wouldn't mess with an OLPF in its current typically fixed position, but is this fixed idea likely to change?

Although Nikon D800E didn't exactly remove OLPF, but negated it, I've found it to be quite manageable; not to mention the footage I have seen with other higher end camera's OLPF's removed.


I'm not preaching any of this, and I know there can be a world of hurt out there for nonchalant use of this practice: I'm just curious about the ( sanctioned) possibilities.
Click to expand...

The image can look OK (if you like fake sharpening, and you like making your actresses' skin look worse than it is - I don't) until you film something that will moire - then everything falls down. I firmly believe a camera is not great because of the few times it gives you an image you can like, a camera is great because it gives you what you want, reliably, EVERY time.
 
Robert, there is a big difference between fake sharpening and what I've been doing, for example. ( if you were referring to that )

But I am doing very tedious processing ( takes > 1 min per frame ) and I am no doing glamour shots, that's for freaking sure.
Absolutely not for casual users. Not ready for prime time.
In fact, I'm not so sure any actress would like the sharp 4K 'look'. Too revealing. I think they like 1K cameras, and vintage low MTF glass.
 
A bit late tonight, will have to get back to it, but I might not be able to comment too much. I ussually only tell about things when they become public from another source, or I no longer need them (what is the use of the inventor giving everything away, when he comes to make his own product and cant bring new features of his own and can't bring his own product to market to afford to do more, because all the competition is using his inventions for free. I know some around here think you should give everything away, but that suites those on top of the heap, and keeps people from making progress for society). I've worked on this in ancient times.
 
Wayne, I'm not sure what you're not understanding about sampling theory. Having 100% fill doesn't make the limits of sampling theory magically go away, and you will still get aliases. Take a look at some of the Foveon images that show strong aliasing. I guess the issue is that some Foveon supporters came up with the claim that their high fill factor obviated the need for an OLPF, but the results demonstrate otherwise.

Graeme
 
Robert, it's true, what me and Les are talking about (I gather) is not artificial sharpening, but restoring authentic like detail. It can be done. Even if you can not restore to more accuracy than it is, you can put something there that looks better. It gets very complex.

Les, 1 min, that's good, the Intel top of the line uprez algorithm, took 100 hours per frame around 10 years ago.
 
Graeme Nattress said:
Wayne, I'm not sure what you're not understanding about sampling theory. Having 100% fill doesn't make the limits of sampling theory magically go away, and you will still get aliases. Take a look at some of the Foveon images that show strong aliasing. I guess the issue is that some Foveon supporters came up with the claim that their high fill factor obviated the need for an OLPF, but the results demonstrate otherwise.

Graeme
Click to expand...

As I was saying, depends on what you call a problem. The example you linked to shows the higher non aligned frequency data affecting the pixel to varying amounts above it. That's perfectly natural, not a problem to me and overcomable by software processing if it was. As I wrote (worth the read) foveon is not an ideal example as it is not true 100% fill and near100% at that with a/micro lens that produces problems, so I laid out a question on an ideal used theoretical sensor, removing external mtf as much as practical. I also laid out a question the previous time about what effects. So, is the graphic example in the link, do you have any other things you are referring to?

I can see such aliasing patterns with the naked eye, and looking through screen doors, just natural. I'm more concerned about unrecoverable missing information. I can see that such things are major concerns in Bayer point sampling, but an ideal used sensir would be much better than a foveon example. The olpf in the example is actually destroying information, if you extended it to cover the whole screen, you would get mainly a blurry mess, go too much and the windows on the apartment building start to disappear, instead of having maybe two blended pixels and two window pixels. I've seen car tires go backwards, its not just cameras. So, I'm asking if there is anything mire substantial to worry about, I'm learning here if there is anything I don't know?

For starters, is there color moire on a foveon sensor, and do you mean the linked example as far as lumina moire? See, with my poor memory I can work out things and not know the term, or learn them and not remember the term.


Thanks again Graeme.
 
A quick googling shows up Foveon aliasing examples like this: http://www.modelmayhem.com/po.php?thread_id=755781&page=3

"That's perfectly natural, not a problem to me and overcomable by software processing if it was." - yes, it's the natural occurrence of a violation of sampling theory - and it's not post-removable.

Graeme
 
@Wayne,

If you sample an audio signal at a certain rate - there is a maximum frequency you will be able to record & replay.

If the signal contains higher frequencies than the rate you were sampling at, your recording will have remnants of this higher frequency data - and when played back, will appear as lower frequency than the original, 'messing up the recording.

Some would argue there is more 'detail', others would explain there is aliasing present.

http://en.wikipedia.org/wiki/Nyquist_frequency

AJ
 
Graeme Nattress said:
Wayne, I'm not sure what you're not understanding about sampling theory. Having 100% fill doesn't make the limits of sampling theory magically go away, and you will still get aliases. Take a look at some of the Foveon images that show strong aliasing. I guess the issue is that some Foveon supporters came up with the claim that their high fill factor obviated the need for an OLPF, but the results demonstrate otherwise.

Graeme
Click to expand...

Good example of this was some of the early Di's (film to digital) didn't use a low pass filter even though they had a rgb sensor per pixel - and they saw a doubling of film grain if the grain size was close to the dimension of the sensor. This is why a 4k di is a bit softer then the raw film (now they either have a low pass filter or slightly defocus the lens, there is btw a weird algorithm in if they do know the exact grain size they can somewhat fix this so they don't have to rescan ... my gut feel is they need to go back and rescan all the film at 6k with a low pass filter so they get clean 4k wavelets).
 
Graeme Nattress said:
A quick googling shows up Foveon aliasing examples like this: http://www.modelmayhem.com/po.php?thread_id=755781&page=3

"That's perfectly natural, not a problem to me and overcomable by software processing if it was." - yes, it's the natural occurrence of a violation of sampling theory - and it's not post-removable.

Graeme
Click to expand...

Graeme, thanks, but that is not what I was asking. The foveon is far from an ideal example, again, you will pickup too many issues from it not being ideal to show the extent of effects on an ideal sensor. Which is the basic start of all discussion, then work from the ideal to what can be done, and the effects of that. Why avoid dropping foveon to discuss it? Foveon like sensors should only get better, so the present is not worth making finale conclusions on (unless it already answers it).

Much of the time you can restore the data, it is just that modern thinking has not caught up to it. I mentioned before, and Les cottoned onto it.

In the end, if a minor olpf has to be used, then OK.

Looked at the example, not the best, the d9 is even less ideal then the latest, and definitely in need of a olpf. But it is what I am saying, it can be restored, somewhat, but without 100% fill there will be gaps in information and twitter to contend with, but still overcomable, however there is no guarantee, but probability depends on the data. However, I do not wish to discuss my solutions for obvious reasons, I am continually developing IP, and have a tool set that can solve these issues, but am checking for more examples. But I wish to discuss realistic degree of issues on an ideal sensor. After we determine the ideal, we can determine what Foveon can achieve. Is that alright Graeme?

I also notice the peppering with the old example, rather than current, and the ones talking real figures.


Thanks.
 
A perfect sensor matters not - it still has to obey sampling theory. You will still get aliases that are not post-correctable.

Graeme
 
The real issue is probably that our vision is so fine we don't notice it much. Like me seeing the inter spacing between 4k pixels, I had to line up the optimum 2400dpi spot of the eye, and consciously look for it, even then I saw a pattern instead of points I could isolate. Normally this is out if focus and also blended neurologically, meaning that unless you isolate it, it probably goes largely unnoticed. It is natural, and as Patrick said, and I've said for many years to get over aliasing, you could defocus the lens a little.

Of course, in sampling theory you double the sampling rate.


Antony, I will have to get back to it. It is late, and finally today I got a oled tablet to enjoy. It is seldom a matter of the problem we know, but the solution. I'm not trying to break mathematics, but to solve the issue. As for a 100% sample area affects, they are what the area, that I accept, but on the examples so far, not as bad as the gappy foveon sensors, given in examples. But people are registering that the problems are getting better the closer they get to ideal, the matter is how close you can get and what can be done about it then. Soon night time about 1 micron sensors should be possible, technology is out there, which will alter the practicalities of non olpf use (and likely start bringing in a heap of other issues).

The stuff about each color layer being a seperate pixel is bad in the example thread, I agree, under this a pixel is a pixel, and performance should be be judge on how well it does on these pixels (which puts Bayer as a disadvantage).



Thanks


Wayne.
 
Graeme Nattress said:
A perfect sensor matters not - it still has to obey sampling theory. You will still get aliases that are not post-correctable.

Graeme
Click to expand...

That's what I've been asking, what are the severity of examples like on the perfect sensor? If you can show me there are severe enough examples , I'm willing to accept that an olpf is a good idea. However, as I said, certain things are acceptable depending on the degree of the effect. However, I know such affects are post rectifiable., maybe you don't know how, but I do. However, I have been explaining everything to clarify the questions, to make it easy, but getting different explanations, have you reading it?

Thanks again.
 
Take a high resolution image and box-filter downsample it. If you know such effects are correctable then you just have to show it.....

Graeme
 
Since we're talking about OLPFs and aliasing, I have a question about CFAs for you Graeme.

How does the choice of CFA affect the resolution of the sensor?

For instance, I stumbled across a nice, albeit low-tech, site about a demosiacking algorithm that a company provides: http://imagealgorithmics.com/RGB_CFAs_Tested.html#RandomRGB

They've tested it on a number of known CFAs and show alternative algorithms. The link I provided goes directly to the application of their algorithm to a "Random RGB CFA" which they state provides superior image quality.

The site also has the test images they've used for download.
 
Graeme Nattress said:
Take a high resolution image and box-filter downsample it. If you know such effects are correctable then you just have to show it.....

Graeme
Click to expand...

Avoidance is not an answer. Ill take it that there are no other types of examples to consider. It's not about an image, that is missing my point (although, generally, you can correct a limited amount in an image). I'm interested in what can not be corrected in footage.

I'll also take it that the industry as a whole generally lacks knowledge the solution and it's engineering.

Thanks for your time and help Graeme.
 
You must log in or register to reply here.
Back
Top