OLPF-mod/fish-net/back-filter for Moire?

Barry_Green said:
Even so, I think the idea of a fully-reliable 1/2" alias-free and moire-free 1080p camera would certainly appeal to a lot of folks, at $799! And if it was switchable, so it could see the full sensor for shallow DOF, and the 1/2" (should we call it 5/8"?) mode for flawless footage... that would be pretty compelling.
Click to expand...

Seems like it would be pretty easy for Canon to implement this. They just need to modify the VGA crop code to do 1080p instead. Or even 1080p and 720p. And have the cropped region movable within the full sensor. Unless the crop mode is somehow hardwired into the sensor
 
shaocaholica said:
1080p cropped on 18MP S35/APS-C is:

(1920/5184) * 22.3mm = 8.26mm wide which is a hair smaller than 2/3". Not great but thats like having a moire and aliasing free 2/3" interchangeable lens (albeit goofy mount for 2/3") camera for $800. Or better yet, a full HD 4x crop mode for super tele.

Are you getting a T2i USLatin? Are you in Studio City, CA? I need to sort some stuff out with the day job but I'm a go to get the D30 OLPF swap done to a T2i as soon as I can source one and I'd like to pit it against whatever you've got.
Click to expand...

+1 on the telephoto, but it is a 480p mode not HD

Yes, in LA. If I have time to spare I am game. I got a 7D, a 24-70mm f/2.8L and a Tokina 11-16 f/2.8 (which would be fun to shoot a brick wall with).

Barry_Green said:
Well, it's actually a lot smaller than 2/3". 2/3" = 9.6 x 5.4, which is 51.84 square mm. Your proposed 8.26 x 4.64 yields about 38.33 square mm. 51.84 is a little bit more than 33% bigger than 38.33.

So the difference is pretty substantial, but still in the ballpark. That 8.26 is closer in size to a half-inch sensor than a 2/3" sensor. Even so, I think the idea of a fully-reliable 1/2" alias-free and moire-free 1080p camera would certainly appeal to a lot of folks, at $799! And if it was switchable, so it could see the full sensor for shallow DOF, and the 1/2" (should we call it 5/8"?) mode for flawless footage... that would be pretty compelling.
Click to expand...

+1 on the price

shaocaholica said:
Seems like it would be pretty easy for Canon to implement this. They just need to modify the VGA crop code to do 1080p instead. Or even 1080p and 720p. And have the cropped region movable within the full sensor. Unless the crop mode is somehow hardwired into the sensor
Click to expand...

I am not waiting on Canon for any sort of updates. They really hurt their machine-gun release schedule. I think their (single?) firmware update several months back was probably just to make us thing that they were adopting Red's business model, and now that they sold several thousand cameras more than they would have otherwise thanks to it they are likely going to wait a year till the next "freebie". BTW, wtf happened with AGC from Magic Lantern? :(
 
USLatin said:
Yes, in LA. If I have time to spare I am game. I got a 7D, a 24-70mm f/2.8L and a Tokina 11-16 f/2.8 (which would be fun to shoot a brick wall with).
Click to expand...

Want to do it today:cheesy:? I found a few T2i body only in LA. Could have it by lunch, test then send off for OLPF swap.
 
Which it could be hard coded into the hardware. Unfortunate for us, you can't really get any schematics of their circuitry :) Believe me, I've been looking.

Mainly because in my research I've been reading quite a few white papers on CMOS sensor design, and how some techniques use more than one active pixel at each location in case one of the pixels fail (and I have seen this with one persons 7D were they had a bad pixel).

Take that a step further, I started thinking, if you are going to line skip in your array anyways, then why not take that dual pixel technology a step further and have some interconnect between the pixels of the rows that is activated in these modes so their stored values actually propagate to the pixels that will be read. Basically, as simple as a switch with a MOS device. But then, speculating with that and the added circuitry (which would be small, but given any extra circuitry eats into the active pixel space), I think the better solution is finding better ways to read the sensor faster. Still, could be useful if done right. But given the debayer pattern, which pixels becomes a lot more complicated since only like colors would be able to be added in this manner.

Given Canon's track record on releasing new features, I'm not holding my breath. Hopefully some more reverse engineering over at Magic Lantern will happen :)
 
USLatin said:
BTW, wtf happened with AGC from Magic Lantern? :(
Click to expand...

Must be a lot more challenging than they thought. Given they are reverse engineering everything from nothing. I've been waiting for it myself but do have a Beachtek that fakes out the AGC circuits. Not the best solution, but honestly, it works well enough for now.
 
I was also thinking that if the 7D/T2i is skipping every 8-9 rows, instead, you could just bin those pixels so your readout is still the same bandwidth (1080p) but you'd be doing a full sensor sample. Downsampling at the pixel level in a sense. Of course this would require more circuitry on sensor (or at the A/D converter) and I'm not sure how that might compromise the rest of the system.
 
shaocaholica said:
I was also thinking that if the 7D/T2i is skipping every 8-9 rows, instead, you could just bin those pixels so your readout is still the same bandwidth (1080p) but you'd be doing a full sensor sample. Downsampling at the pixel level in a sense. Of course this would require more circuitry on sensor (or at the A/D converter) and I'm not sure how that might compromise the rest of the system.
Click to expand...

Yep, that was essentially what I was sort of referring to. I'd say he'd need to be something that happened in the pixel array since I think the sensor might not be fast enough to read all all the pixels even to the A/D?

That big problem is we don't know where the slowdown is in the system. Is it the actual reads from the pixels? Is it the A2D on each row? Is it the data bus in the system? Given the camera can do 8 raw frames a second, is it storage that is a problem?

I'm thinking it is something between the pixel read speed and the A/D given the skew/jello of the camera. The time between the first and last read is pretty long since they employ a rolling shutter read technology (like most CMOS sensors do).

The complexity also is as I said with the debayer pattern. Two rows do not have the same pixel colors at the same corresponding position, so you'd need every other row pixel to be binned (if following the standard more green than red or blue debayer). I could see the circuit complexity getting out of hand quickly.
 
Just picked up the T2i 30min ago. Sorry USLatin but this thing is going straight to WA state for the OLPF mod. Gotta get things sorted and in the mail by the end of the day. Hopefully I can get this thing to Barry for SXSW.

Edit:
Camera is on its way! Will be back in my hands by the end of next week. Fingers crossed.
 
Last edited:
From another thread:
http://www.dvxuser.com/V6/showthread.php?p=1906723#post1906723

Jordan_Sopinsky said:
Won't applying a stronger low pass filter make the image blurry? It's the aliasing that makes the image seem sharp. Yeah, you'd get rid of the aliasing, but it would be at the cost of "full"
HD video.
Click to expand...

Well, the current low pass filter in the 7D/T2i is designed for 18MP which is way way more than 1080p. All that detail above 1080p is useless in video mode and adds to the moire/aliasing problem. The new filter I'm having installed is rated at ~2K or slightly over 1080p. It will for sure make the 18MP stills blurry but the hope is that the video won't lose any resolution but the moire and aliasing will be greatly reduced.
 
Last edited:
Everything is all figured out. The camera is en-route for the swap and I've made accommodations to get it into Barrys hands ASAP after I get it back.
 
Hope the mod works out ok, might make a big difference. Is it reverse-able? Does Barry have an unmodified 550D in order to test before and after ?
 
I think it is going to limit the amount of moire to a smaller range of frequencies, but the problem will still be there on detail equal to lines only resolved by 1080p, and not by 720p. Which makes me wonder if it will do anything at all.

Right now I would guess/bet the only mod worth doing would be a custom ~720p OLPF with IR.

This 2k OLPF is however a great start, since it is totally worth finding out if any positive impact can be had without losing detail.
 
Last edited:
Minor update. Camera has arrived safely at the shop and the swap should be done by tomorrow and shipped back to me tomorrow as well. They have a 3 week delay right now so I had to pay for rush but they understand why I need this done ASAP so Barry can take it to SXSW.
 
Updates:

D30 OLPF installed sucessfully today
It fits physically
Auto focus recalibration for stills successful
should be sent out today, overnight service
 
Man, can't wait. Did you shoot something before you sent it so that you could shoot it again that has moire/aliasing? :)
 
You must log in or register to reply here.
Back
Top