Something isn't right.

[dinzitari]

So, I've read through this Anamorphic guide, which was helpful in many ways. Still, i'm having trouble understanding this "stretch" method of supposedly gaining additional resolution. How is it possible to output 2k/4k, when your actual recording is 1080p? How can we output to a theater in this "phantom resolution" when blu-ray is maxed at 1080p? 

 

This entire concept seems mathematically and logically erroneous to me. Someone please prove me wrong, because i'd love to be wrong!

[Caleb Genheimer]

By stretching horizontally, you don't loose any of your 1920x1080p pixels, But the 1920 are doubled (2 of each next to each other in place of one). 

 

Technically speaking, although the frame size becomes 3840 pixels wide, half of those are duplicates (or more precisely, interpolations) of the original 1920 across. 

 

BUT (and here is the genius of it), you do not loose perceived sharpness because the 1080p vertical lines of resolution are all still there.

 

One last trick that can seal the deal is after the stretch out, you apply a grain overlay at the increased resolution. I've done a few things on GH2 with my Kowa 2X that I have stretched out, cropped (to get 2.35:1), and grained. I can confirm, it looks like real-deal 2560x1080p. If that ain't 2K, I don't know what is.

 

This mostly works because of the maintained 1080 lines of resolving power. That keeps things looking crisp and sharp even when you stretch out the other dimension. It also works because 1080p and 2K aren't really very different, they're pretty close in terms of resolution. Mostly the anamorphic aids in changing to the wider aspect ratio while maintaining 1080p vertical lines of resolution.

[dahlfors]

Many cameras do vertical line skipping on the sensor to achieve 1920x1080 resolution, due to limited in-camera processing capabilities and how the sensor reading works - from top to bottom, line by line.

 

This means that each horisontal line has the full line readout to use for downsampling, which if the in-camera processing is maintaining quality good enough should result in cameras having better quality/resolution on the image horisontally than vertically (where resolution & data gets lost due to lines being skipped).

 

This is good for anamorphic use, since we stretch out the footage where there is most of the resolution & detail. If the upscaling interpolation is done right and with a bit of grain as Caleb mentioned - you will perceive the resolution as being better.

 

If you compare an upscale from 2x anamorphic, 1920x1080 -> 3840x1080, the result of that will be perceived as having much more resolution than an upscale from 1920x1080 -> 3840x2160 - since then your footage has been scaled in two dimensions, which makes it much more noticeable.

 

"Gain resolution" is not the best term in my opinion. Closer to the truth is: "you perceive it as much more detailed even when upscaling it horisontally, since the vertical resolution is still full res".

 

For Blu-ray output, you'll have to squeeze your footage vertically instead.

 

If you want higher output for theathers, your best option would be to deliver as a digital cinema package. Some more info about that available here: http://nofilmschool.com/2012/07/project-in-digital-theater-make-digital-cinema-package-for-cheap-with-opendcp/

[Sean Cunningham]

Softness in the horizontal plane of an image, and our visual system's ability to more or less borrow vertical resolution to maintain a pleasing, non-distracting image, has been exploited by video engineers since the dawn of...video.  Anamorphic techniques are one of the only exploits used for an enhancing, aesthetic gain versus a cheap, technical shortcut to saving bandwidth.

[dinzitari]

Many cameras do vertical line skipping on the sensor to achieve 1920x1080 resolution, due to limited in-camera processing capabilities and how the sensor reading works - from top to bottom, line by line.

 

This means that each horisontal line has the full line readout to use for downsampling, which if the in-camera processing is maintaining quality good enough should result in cameras having better quality/resolution on the image horisontally than vertically (where resolution & data gets lost due to lines being skipped).

 

This is good for anamorphic use, since we stretch out the footage where there is most of the resolution & detail. If the upscaling interpolation is done right and with a bit of grain as Caleb mentioned - you will perceive the resolution as being better.

 

If you compare an upscale from 2x anamorphic, 1920x1080 -> 3840x1080, the result of that will be perceived as having much more resolution than an upscale from 1920x1080 -> 3840x2160 - since then your footage has been scaled in two dimensions, which makes it much more noticeable.

 

"Gain resolution" is not the best term in my opinion. Closer to the truth is: "you perceive it as much more detailed even when upscaling it horisontally, since the vertical resolution is still full res".

 

For Blu-ray output, you'll have to squeeze your footage vertically instead.

 

If you want higher output for theathers, your best option would be to deliver as a digital cinema package. Some more info about that available here: http://nofilmschool.com/2012/07/project-in-digital-theater-make-digital-cinema-package-for-cheap-with-opendcp/

Thank you everyone for the insight, I appreciate everyone trying to clear this up for me. 

 

Still, I am confused. Once you capture a pixel, that pixel is that pixel. You can't expand detail or resolution from it, right? So, If we expand the horizontal resolution to create the proper "image ratio/appearance" aren't we still softening? I mean, we can't create more image data, because we didn't capture it. It was captured at 16x9 and shall remain that no?

 

These are pretty advanced topics, since we are now dealing with "tricking the visual perception". Does anyone know of some place I can find detailed info on this?

 

 

[Caleb Genheimer]

Yes, you are softening, but the human eye doesn't perceive it because the vertical resolution has not been softened . . . just the horizontal. 

[Tito Ferradans]

I think the subject here would be he "circle of confusion", right? That's the exact name, if you wanna read it further. it's not just about anamorphic, but image perception by the human eye/brain.

How big can the smaller piece of information be, and still look "sharp"?

[dahlfors]

These are pretty advanced topics, since we are now dealing with "tricking the visual perception". Does anyone know of some place I can find detailed info on this?

 

Tricking visual perception has been done a long time in art, and yes, how the human brain interprets visual information is a big subject.

 

I'd recommend that you experiment in Photoshop with upscaling on some stills (a lot faster and easier to work on a still). Pick some sample stills that you can find, one squeezed anamorphic still and one normal still with no squeeze.

 

Then use these techniques - I often use them myself in my design work if I for some reason have photos that aren't detailed enough:

 

1) Depending on the image, add one round of sharpening with smart sharpen at around 1.5-3.0 pixel radius, with amount around 5-15%,

 

2) Then add a second sharpening with 0.2-0.5 pixel radius at a bit higher amount, 15-45%. The values depends a lot on how it's going to be used, and the original resolution of the image.

 

3) If I want yet more perceived resolution I add one noise layer in overlay mode and tweak that to what I like. Since you do it for screen use, you use quite little opacity on the noise layer (how large of an opacity value you will use also depends on how much contrast your noise layer has). When working with print as output medium - you use far more noise (the print process hides most of the noise you add in fact).

 

 

Experiment with different settings and put the images side by side to compare. You will notice that you can get a lot more perceived resolution when you sharpen the right way, add grain - and do only horisontal scaling as in the case of anamorphic...

 

- Same methods can be applied for video as with the stills - the difference will be that your noise is animated and that you don't necessarily run two rounds of sharpening. These methods (except horisontal-only scaling as with anamorphic) have been used for print medium a long time. For print you can successfully scale upwards to 150-200% without much perceived loss of resolution, even more when it goes for billboards or something that will be viewed at a further distance.

 

- Basically, most kind of grit / noise / dirt / distortion layers layered on top with a bit of opacity will increase the perceived resolution. In the early days of web design when high res digital photos were harder to acquire, people used scanlines so much for this reason - that it became a popular design style (similarily to this image: http://fc06.deviantart.net/fs70/f/2010/081/9/f/Labrador_Wallpaper_Scanlines_by_Red_Se7en.jpg). Scanned filmgrain or noise will be what you want for video.

 

Similar methods for upscaling were previously commonly used for 3D animations to save on render time previously when hardware was a lot slower. Genuine Fractals was the name of a tool specifically made for upsizing, yielding even better results than traditional upsampling algorithms. These days it has been renamed "perfect resize": http://www.ononesoftware.com/products/perfect-resize/, but it's unfortunately not a plugin for video softwares and if it is used on video, it needs to be run frame by frame.

 

With these upscaling methods you'll notice that upscaling can produce very good results. Thanks to anamorphic needing only horisontal scaling, end result is even better.

[dahlfors]

I also mentioned this (and a lot of other perception tricks) has been used in art. You could have a look at these two talented digital painters, who use noisy brushes / texture brushes for adding detail to their images. Even though they use simple brush strokes in certain places, your brain interprets and adds detail to the artworks: http://www.tuomaskorpi.com and http://simonstalenhag.se/