LOOP made from Automechanic "Jenny O"
Hi there boyz and girlz
Over at personal view we – Jeremiah, Sean and I – have been developing a sticky syrup for sharpening footage, though it is much improvable, I’ll like to share a recap of our findings with you =)
The good Jeremiah started all
Which needed a workaround for the find edges filter, in order to use this technique in 16 and specially 32bits mode
So instead of “find edges”, I tried CC Kernel (for adobe cc) or alternativelly threshold or cartoon, yielding each different results. With the kernel 1,2,1 / 0,0,0 / -1,-2,-1 . I tried Luma matte instead of inverted.
Here’s what I did
Created sequence (SH) with Footage (A).
Then duplicated 2 times that footage; A1 (on top) + A2 (below), precompose those copies (B).
Inside precomp B, put filter gaus blur ratio 2 in A2, set A1 to difference. Pushed Levels in both (individually, not smartass adjustment laWyer)
Back to SH, I duplicated A again (A3) and put over A with the sharpness I want to apply, set layer to Luma matte “B” which is on top.
In precomp B put, in this order, the Kernel CC filter mentioned before, plus Levels, Tint and another gaus blur ratio 2. With levels control what the matte filters black = masked OUT , white = masked IN.
Unchecked visual of precomp B. This is the effect. I still precomped B + A3 (C).
Then in master seq (SH) duplicated C two times, and set one to to lighten and the other to darken… changed opacities till reached point of interstellar fusion. That’s it, close enough… more testing required, aaaaahhhghg going to zzsslleeeep now
Quick and dirty video test using method described above (Kernel mask / difference). In this, lighter sharp layer was about 33% and darker about 55%.
Personaly, as I would like to add my own grain recipe to final grade, being able to sharp footage not introducing further grain/noise would be very nice.
Issues to overcome are substituting gaussian blur for something that repeats or feathers pixels on the edge of frame AND simplifying most possible in order for the whole process to be less processor intensive and quicker.
Please, have in mind this footage is not from GH2, but from a 1/2.3” sensor P&S camera 😎
LOOP made from 331Erock "Halo Meets Metal"
Then master Sean Cunningham came to share his incredibly useful and thought out high-pass method… among other goodies =)
With the kernel effect the lines can be very thin and detailed… I needed more time to play more with it. Meantime, there was also the cartoon option, also with normal Luma matte; working pretty well too 🙂
Studying frequency filtration – much used in photography – possibilities for video too. We needed to find a ps “add image” like 32 bit capable AE pluggin.
Like in this video.
Yes it works. Added to Frankensharpen project file 😛 • LINK below
Box blur works the best ’cause you can really further control edges with itinerations… but one would need 3 lives or Deep Blue =)
AND finally the
downloadable project for AE
Corresponding to the first video above (vimeo):
Because differences in how techniques are built, I’ve chosen to “equalize” the kernel part to match HP. No added grain after sharpening, camera GH3, -5-2-5-5, lens Vivitar 28-200 f3.5-5.3 aka Kobori^Shugga
As long as you tweak first box blur, levels and fast blur settings, the kernel method seems good to me IF it could be turned lighter on processing… or you have a powerful render workstation that chews whale bones for breakfast 😛 • As you can see below, lighten layer was set to 55% and darken to 77%. Also in the frame grab you can check how less noisier kernel is (normal as it masks), of course it is a still, you’ll notice less of the grain in motion, but still
BurnetRhoades’ high pass approach is very coherent and much more efficient,
like 4 times faster render, see below, more than good for most uses =)
a peep into the final project layout
• none of this would have happen without Jeremiah or Sean •
• or the hostage of Vitaliy Kiselev •