Enhancing Canon Eos APS-C camera videos

[otherman]

Canon Eos cameras are knowed fo good quality video, but the quality is not so great for post-production. APS-C Sensor camera is the most common, and they need the work the most. I've a 550d, and I'll work on his video files.
My general intent for this discussion it's to find (or to write) the best scripts to get the best from this cameras.

In my idea, there are three areas of interest:

1) Detail enhancement
2) Compression artifacts
3) Color Depth (for color grading)

I will update this first and the second post with all informations and (hopefully) solutions for this theme.


DETAIL ENHANCEMENT

550D, 600D, 650D, 700D, 100D, 1200D, Eos-m and Eos-m2 have all the same APS-C Cmos sensor with 5184x3456 of real resolution. To get fullHD 1080p video, they get every third line (in vertical and horizontal) getting 1728x1152, and crop to 1720x974 that is the real resolution of the videos, then use a fast bilinear resize to final output at 10920x1088 (1.116x).

How can I reverse the fast bilinear resize to get off his softness (to resampling with better algorithm)?

I will soon link a video sample from my camera and edit this post.

[otherman]

Hopefully reserved for solution!

[Lyris]

Canon Eos cameras are knowed fo good quality video,
They are, but IMO the quality is not good at all. I think they're known for being able to give shallow depth of field for low prices. Other than that, the quality is poor.

I've tried to get better results out of those, but gave up. I'd be curious to see if anyone has better responses, but I imagine anything you do to try and get rid of artefacts like moire is just going to blur the video more.

I switched to a hacked Panasonic GH2 camera which gives much better quality and excellent resolution, although not as shallow DoF.

[Reel.Deel]

How can I reverse the fast bilinear resize to get off his softness (to resampling with better algorithm)?


I have yet to try this plugin but maybe this is close to what you're looking for:
debilinear r6.zip (http://rgb.chromashift.org/debilinear%20r6.zip)

From the read me:
This filter is designed to reverse the effects of bilinear upsampling.
Parameters and sampling pattern try to mimic Avisynth's BilinearResize.

And here's a more detailed description by Firesledge (http://forums.bakabt.me/index.php?topic=37168.10;wap2):
Debilinear is intended to reverse the bilinear upscale process. It’s incredibly effective and allows upscaling again with a better algorithm, or leaving the video downscaled and trusting the player upscaler.
But this works only on "ideal" or close to ideal cases. The condition is that the material has to be a pure bilinear upscale, an you have to know exactly what were the original resolution and the possible sub-pixel shifts.
Any other material overlayed on top of the upscale (like native 1080i/1080p credits) will cause ringing or visual defects. Same with compression artefacts, which will be amplified to death, so debilinear is clearly not suited for TV streams.

Hope this helps.

[otherman]

Thank you, I'm making some tests :)