Letus35a Image Sharpening Techniques ???

[Ben Winter]

I tried the technique Frank suggested, and didn't get vastly better results than with just plain sharpening--but they were slightly better. I received some generation loss (weird, faint horizontal red and green lines) when I used the unsharp mask on just the raw footage. Taking the HD res noise (frank's) route and unsharping gave me better results without the degradation and were a little bit cleaner. The Letus GG wasn't shaking enough and left some ugly black grain in the footage, but I corrected the unit with a hands-on fixer-upper so hopefully future footage won't suffer. Here are the results:

Canon GL2
Letus35(quasi-A)
Canon FD 55mm 1.2 @ f2.8
ND filter ON
Camera: f1.8, shutter 1/60, 0dB

Original Footage 1.36MB 15 seconds

Plain Unsharp Mask 1.18MB 15 seconds

Frank's Technique 1.18MB 15 seconds

Ignore the contrast (a byproduct of the unsharp mask) and really look at the details, like the tire tread and the pinstriping on the car. On my screen the regularly sharpened footage was more blotchy. The noise-sharpened footage does have a louder ting of precision to it. Remember to look for sharpness only in the middle of the in-focus area ("that background ain't sharp at all!"--you know what I mean though, your eyes can wander) What do you all think?

[Frank Ladner]

Hey guys!


Just realized that above I said 1920x1080. When using 720x480 NTSC, you should use 1620x1080 (1920x1080 is wide / HD 16:9).

Ok, I have put some test images up on my site:

This is just showing what EE can do to a low-res 720x480 image:
http://www.frankladner.com/images/sharp_01.jpg


Original 720x480 frame:
http://www.frankladner.com/images/sharp_frame.jpg

Closeup with grain added (no sharpen filter yet)
http://www.frankladner.com/images/sharp_closeup.jpg

Original with closeup:
http://www.frankladner.com/images/sharp_original.jpg

Sharpened at 720x480 rez with closeup:
http://www.frankladner.com/images/sharp_720x480.jpg

Sharpened at 1620x1080 rez - grain added - with closeup:
http://www.frankladner.com/images/sh...edandgrain.jpg

I suggest opening the last three images in separate tabs/windows and flipping between them to see the difference.

For instance, flip between sharp_original and sharp_720x480, then between sharp_original and sharp_enlargedandgrain, while looking at the squirrel's eye. Flipping between the last two frames will reveal that the sharp_720x480 looks a bit more blotchy.


I took the original 720x480 footage, dropped it into a 1620x1080 composition, and scaled it up to match the composition size. ( 225% )

Comp 1 = Original Footage (720x480)
Comp 2 = Comp 1 scaled to 1620x480
Comp 3 = Comp 3 with grey solid layer (w/ Noise applied) on top with Overlay mode

The Unsharp Mask filter settings were identical for the original 720x480 frame, and the 1620x1080 grained frame.

Now, this is just dealing with a single frame. If you don't like the way AE's Noise looks in motion, there are other plugins that may work better, such as CineLook.

Also, as with other things, add Grain/Noise to taste. The main thing is not so much to get the grain effect, but to give the sharpen filter finer areas to work on, as opposed to "blocks" of identical-colored pixels.

I will say that for some, this may be more trouble that it's worth, and that just applying the Unsharp Mask to the original footage may be all that's needed. However, be careful when sharpening footage that was also sharpened/enhanced in-camera as that will bring out the fringing around contrasty areas.


I hope this makes sense. Please let me know if you guys want more clarification.

[Jamie Niebuhr]

neato, the footage definitely looks sharper with your process...

[Frank Ladner]

Thanks for the feedback, Jamie!


I keep thinking about whether or not there's a better way to approach this than using random noise.

I mean...the point is to break up the 'blocks' and give the sharpen filter finer areas and higher resolution to chomp on.

...but there has to be a way, programatically, to "split" the blocks of pixels using some interpolation, as opposed to adding variances in color to the blocks. I'm thinking of a way to blow the image up, AND re-introduce detail based on the image, not on random noise pixels.

For instance, when sizing images down, for each destination pixel you take the neighboring pixels (8 neighbors around that pixel, for example) and get the average of their values. This gives you the new pixel to plot in the new, sized-down image.

NOW...how can that be reversed? Well, you've lost information that you can't accurately get back when enlarging the image. I imagine you can super-sample the areas around a pixel and for each joining pixel, add a pixel of new value between them based on the collected information about the surrounding pixels.

I am a simple-minded person and don't know much about how B-Spline, Lancoz(sp?) and other algorithms work with resizing images...but I have an idea of how this information can, theoretically, be "brought back" from the smaller image. Easier said than done, though.

Well, I'm just rambling on but I think there's a way to get sharp and high-rez looking footage from regular SD. (Given that care is taken while shooting it - such as avoiding sharpening too early.)

[Michael Maier]

Originally Posted by Ben Winter

I tried the technique Frank suggested, and didn't get vastly better results than with just plain sharpening--but they were slightly better. I received some generation loss (weird, faint horizontal red and green lines) when I used the unsharp mask on just the raw footage. Taking the HD res noise (frank's) route and unsharping gave me better results without the degradation and were a little bit cleaner. The Letus GG wasn't shaking enough and left some ugly black grain in the footage, but I corrected the unit with a hands-on fixer-upper so hopefully future footage won't suffer. Here are the results:

Canon GL2
Letus35(quasi-A)
Canon FD 55mm 1.2 @ f2.8
ND filter ON
Camera: f1.8, shutter 1/60, 0dB

Original Footage 1.36MB 15 seconds

Plain Unsharp Mask 1.18MB 15 seconds

Frank's Technique 1.18MB 15 seconds

Ignore the contrast (a byproduct of the unsharp mask) and really look at the details, like the tire tread and the pinstriping on the car. On my screen the regularly sharpened footage was more blotchy. The noise-sharpened footage does have a louder ting of precision to it. Remember to look for sharpness only in the middle of the in-focus area ("that background ain't sharp at all!"--you know what I mean though, your eyes can wander) What do you all think?

What's the dark fixed pattern over the image? It’s in the original clip as well.
The out of focus background has lots of aberrations and they get worse with the artificial sharpening.
In my opinion it looks like sharpened video. Not sure I would use the process for a serious production, but thanks for taking the time to do it and for sharing.

[Ben Winter]

That's the Letus ground glass not moving very far. I've since fixed it. Where's this aberration you speak of? I see none of it.

[Frank Ladner]

Ben: Nice footage! I see what you're saying about the Letus ground glass. It seems it doesn't produce any noticeable grain, but instead can sometimes make these dark, transparent spots (ghosts) from the grain moving in circles. I haven't really noticed it in my footage, but I can point the device at a bright light and can see the pattern (seem like my eyes have gotten trained to spot those kinds of things from all my messing around with microcrystalline wax).

[Ben Winter]

This is what I did to fix it: I pulled out the little rubber stoppers that hold in the white plastic sticks a little (about a millimeter) which allow the sticks to move around a lot more. A millimeter makes a noticable difference and you can pull it out even more if you want. Then, I took a glue gun and heated up the glue that was keeping the other ends of the sticks on the GG "plate", if you will, and pushed the GG down about a millimeter to compensate for the change in distance.

I don't know if the issue is with a few units or with a lot of them, but it sure is a shame the Letus has this caveat out-of-the-box.

[Frank Ladner]

Thanks a lot for the tip, Ben!

I'm likely going to take my Letus35 apart for a good cleaning soon, and I'll try your adjustment.

Also, have you tried higher-powered batteries?

I do a lot of outdoor stuff with a high shutter speed, which could actually freeze the grain pattern in the footage. Maybe if I can get it to wobble a bit faster, that would do the trick.

HOWEVER, I don't know what these little motors are rated for.

[Cody Dulock]

mark it looks like the focus on your camcorder isnt set correctly or your letus35 flange focal length isnt correct or the GG is wobbling a bit. these are the most common problems that can occur.

[Ben Winter]

I talked to Quyen a while back and he seemed to think 3 AA's weren't too much. So 4.5v is at least OK. I'm not sure if it would help things though--makes me think it would just keep the GG moving in a smaller circle (follow my logic here: if the weight in the motor swings in one direction while the momentum of the GG is still going in the other direction, they essentially are going to cancel each other out or at least make for a smaller movement on the next oscillation.) Therefore, there has to be a "sweet spot" vibration movement where the heave-to and heave-from of the motor perfectly align with the inertial tendency of the vibrating GG plate.

Think of it like swinging on a swingset; you always swing your legs forward when you're about to go forward, and backward when you're falling back to augment that motion. It doesn't make sense to swing your legs forward as you're going backward; that will only slow you down. I'm going to ask Quyen for the specifics of the motor so I can buy the proper potentiometer. I think being able to finely tweak the speed of the motor will be a very powerful tool in moving the GG the maximum distance. Thoughts?

EDIT: I also tried 3 AA's on the earlier model of the Letus, for anyone who's interested. It definately helped things, but the plastic sticks that hold the GG and allow it to move were much longer which meant a looser vibration and more movement. Shorter sticks mean it's going to be trickier.

[Mark Wisniowski]

Hey Ben and Frank, thanks for taking the time out testing this technique !

[Frank Ladner]

No problem, Mark! Sorry for the late reply!

Has anyone else tried the technique?

I wanted to add something else. When shooting with the GL2 and sharpness turned down, it is quite difficult to know for sure when something is in focus. So what I do is use the CP (Custom Preset) button to toggle on/off the sharpness setting. I do the focusing while sharpness is normal and before recording, switch it off.

(Also, I have the color saturation turned down a bit also.)

[Leo Mandy]

Great idea Frank! I like the footage without the grain though, the grain is a little too much for me.

Ben : remember when pulling out those little stoppers from the unit that Quyen uses a little device to measure the entire area of the plastic mount for the GG with - if you put a stopper in too far on one of the legs, the FL of the GG is going to be off on one side of the screen!

[Daniel Rudd]

Anybody making gg rotational adjustments on one of the new flip models?

I'm having the same problem that it soudns like Ben is describing.

But the new flip model is much more heavliy contained. I hate to open it up.

I'm having a silmilar issue to this clip posted by Mark.
notice the grain patterns moving through the dark areas on the hood.

http://www.probe3.com/3ccd/Letus35a/L35A_wrx_side.mov

Does this have more to do with my focus, or with my gg range of motion?