I recently got turned on to a fantastic site with Photoshop tricks at http://luminous-landscape.com/tutorials.
In particular, I found the sharpen tutoral at http://luminous-landscape.com/tutorials/high-pass-sharpening.shtml gives an approach far superior to the Sharpen or UnsharpMask filters. Better control and reduced "ghosting" around the edges, without the usual heavyhanded/dull contrast.
So last night I was inspired to try and duplicate it in VV.
First, there is no HighPass filter in VV, however I was eventually led to the Convolution filter as a rough alternative.
What a wild effect. After much cluelessness I eventually tried:
-1 -1 -1
-1 10 -1
-1 -1 -1
in auto-normalized mode which did the trick with some "side-effects" -- for certain values of the middle value, all the edges go white.
I was about to move on when I checked the drop down templates.
Duh ;-) Hidden there is a "Sharpen" template, and it explained my side-effects.
(Don't use the same -1 values on all outside cells because it just cancels out, rather than enhancing differences. Look at the values in the VV templates.)
Ok, so armed with the VV sharpen template in the convolution filter, I moved on.
(It would probably be useful to add a threshold value to the filter. This might allow one to control adding too much noise in the clear backgrounds. AFAIK, it is impossible to simulate threshold strictly with values in the convolution filter.)
The photoshop tutorial suggests a "Hard Light" blending between layers. I have no idea what this does or what the equivalent might be in VV. So I just ignored it and used the usual opacity level in VV.
So if you haven't looked at the photoshop tutorial, the procedure is:
1) Duplicate your video track.
2) Apply the "Sharpen" template of the convolution filter to your front track.
3) Blend the two tracks by reducing the opacity of the front track.
In short, the effect is stunning. It doesn't work for every scene, but I am blown away when it does. I have a clip where I was walking in a shallow lake with the camera facing down. The effect makes every little ripple jump out and the sun glints. The sharpen/unsharp mask filters increase contrast harshly and add no detail. The convolution filter shows ripples I didn't even know were there :-) And it looks real. I can crank the opacity up to 100% and you still might not know it.
I also applied the effect to some VHS transferred video (taped from an SVHS deck to my TRV-900, then firewired into the computer). I had done this as a test and was about to give up because the quality was poor and the sharpen filters were too obvious. The convolution filter simply nails it. And, the convolution+broadcast-clamp effects render to mpeg2 at least 3 times faster than the unsharp mask filter.
Well, I'm excited enough about the result to risk posting something that others already know and have mentioned before. And if SoFo posts that "well the sharpen filters are simply convenience filters on top of the convolution filter" then I'm going to be embarrassed ;-)
I'd be interested if anyone knows the math behind the Photoshop Highpass filter. Perhaps I've missed an equivalent VV filter, or perhaps someone will be inspired to write one. And perhaps there are other ways to blend the two tracks that give extra control.
In particular, I found the sharpen tutoral at http://luminous-landscape.com/tutorials/high-pass-sharpening.shtml gives an approach far superior to the Sharpen or UnsharpMask filters. Better control and reduced "ghosting" around the edges, without the usual heavyhanded/dull contrast.
So last night I was inspired to try and duplicate it in VV.
First, there is no HighPass filter in VV, however I was eventually led to the Convolution filter as a rough alternative.
What a wild effect. After much cluelessness I eventually tried:
-1 -1 -1
-1 10 -1
-1 -1 -1
in auto-normalized mode which did the trick with some "side-effects" -- for certain values of the middle value, all the edges go white.
I was about to move on when I checked the drop down templates.
Duh ;-) Hidden there is a "Sharpen" template, and it explained my side-effects.
(Don't use the same -1 values on all outside cells because it just cancels out, rather than enhancing differences. Look at the values in the VV templates.)
Ok, so armed with the VV sharpen template in the convolution filter, I moved on.
(It would probably be useful to add a threshold value to the filter. This might allow one to control adding too much noise in the clear backgrounds. AFAIK, it is impossible to simulate threshold strictly with values in the convolution filter.)
The photoshop tutorial suggests a "Hard Light" blending between layers. I have no idea what this does or what the equivalent might be in VV. So I just ignored it and used the usual opacity level in VV.
So if you haven't looked at the photoshop tutorial, the procedure is:
1) Duplicate your video track.
2) Apply the "Sharpen" template of the convolution filter to your front track.
3) Blend the two tracks by reducing the opacity of the front track.
In short, the effect is stunning. It doesn't work for every scene, but I am blown away when it does. I have a clip where I was walking in a shallow lake with the camera facing down. The effect makes every little ripple jump out and the sun glints. The sharpen/unsharp mask filters increase contrast harshly and add no detail. The convolution filter shows ripples I didn't even know were there :-) And it looks real. I can crank the opacity up to 100% and you still might not know it.
I also applied the effect to some VHS transferred video (taped from an SVHS deck to my TRV-900, then firewired into the computer). I had done this as a test and was about to give up because the quality was poor and the sharpen filters were too obvious. The convolution filter simply nails it. And, the convolution+broadcast-clamp effects render to mpeg2 at least 3 times faster than the unsharp mask filter.
Well, I'm excited enough about the result to risk posting something that others already know and have mentioned before. And if SoFo posts that "well the sharpen filters are simply convenience filters on top of the convolution filter" then I'm going to be embarrassed ;-)
I'd be interested if anyone knows the math behind the Photoshop Highpass filter. Perhaps I've missed an equivalent VV filter, or perhaps someone will be inspired to write one. And perhaps there are other ways to blend the two tracks that give extra control.