Post sharpening on shallow DOF shot?

G

Gene Sung

Veteran
I haven't got a 5D3 yet since I'm waiting a bit too see how the upcoming Sony pans out before I make a final decision, so I cannot test this myself.

I've seen Philip Blooms video where the post sharpening does wonders to the footage. HOWEVER, all the shots were deep DOF, wide landscape type shots.

How does the post sharpening work on Shallow DOF shots like portraits or more close focus type stuff? Or night shots where there is a lot of Bokeh from car lights?

Logic says that post sharpening and shallow DOF might not pair as well as landscapes and deep DOF. Especially on shots with blurry backgrounds or night lights, it seems like post sharpening would make a mess of the Bokeh.

So what have your experiences been with post sharpening more shallow DOF shots?
 
I've shot some night street, car light stuff and post sharpened, looks great. Shot shallow DOF too, no issues. The sharpening is very slight and doesn't mess with the image just brings out more detail same as sharpening a RAW still after converting it to a jpeg. Here's a resized screen grab straight out of the MKIII IPB 2500 ISO 0 sharpness.

2500 ISO copy.jpg
 
Will Turner said:
Nice and detailed but the halo's caused by the post sharpening are just grim in my opinion.
Click to expand...

Zero halos when in-camera sharpening is set to 0 and post sharpened (halos come from Unsharp Mask style sharpening with a radius > 1. Won't happen with a convolution filter and less likely with USM and radius < 1).

In-camera sharpen 0 and significant post sharpening (medium DOF, background is blurred slightly):

In-camera sharpen 0, varying amounts of post sharpening on difficult material; still very little aliasing (Bugatti grill is only area noticeable if looking for it). Many shots with the somewhat soft 16-35 F2.8L II. The 70-200 F2.8L II shots required less sharpening (one of the sharpest zooms available: matches the sharpness of decent primes).

I have many narrow DOF shots for my movie- post sharpening worked well (no halos or other weird artifacts).
 
Do you guys sharpen in post with 5DII footage? Why is post sharpening so common on 5DIII footage? Is it the line skipping or something?
 
jkim3381 said:
Do you guys sharpen in post with 5DII footage? Why is post sharpening so common on 5DIII footage? Is it the line skipping or something?
Click to expand...

At 0 sharpness the MKIII is a little softer than the MKII. Turning up the sharpness in camera produces a halo effect. Post sharpening produces excellent results. I suspect the softness is due to a denoising algorithm that affects all ISOs and can't be turned off.
 
What sharpening tools are you using and in what post apps? FCP, Resolve, Premiere, After Effects, etc...
A few common settings would be nice to know too...

@JCS
A brief explanation of "convulution" sharpening and which tools offer it would be helpful.
We should sticky it over at C5D as well...
 
Last edited:
So far I'm just using the basic sharpening filter in FCP7 and adjusting the setting to taste (around 20). There are 2 sharpening filters in FCP7, one of them is really ugly.
 
The 2nd one in the sharpen menu is the good one (only has amount adjustment). The first sharpen filter has controls for intensity and amount, no matter how much I make adjustments it looks plasticy.
 
In PPro, convolution sharpen is: Effects/Video Effects/Blur & Sharpen/Sharpen (GPU accelerated). Unsharp Mask is below it.

Convolution Sharpen works by starting with a center pixel, and sampling surrounding pixels in a square (typically forming a 3x3 grid). 9 pixels are sampled and multiplied by positive and negative numbers ("weights") in a matrix corresponding to pixels, and the resulting values are added up for the final value. This is done 3 times, once for R, G, and B (or Y, U, and V, depending on implementation). It won't form halos, even at high amounts, but will show artifacts and aliasing at high amounts.

Unsharp Masking (USM) works completely differently. USM first creates a blurred copy of a region of pixels, then subtracts the blurred pixels from the original pixels (and rescales them so brightness is not affected). With a radius of < 1, it looks similar to Convolution Sharpen. With a radius of >=1, it can form halos (as in-camera sharpening does on the 5D3). With a really large radius (50-100+), it can be used to perform Local Contrast Enhancement. While USM could easily be GPU accelerated, it's not in PPro (including CS6). Too bad PPro can't run custom pixel shaders as can AE. This would allow many custom, real-time GPU accelerated effects to be performed in PPro that aren't built in.

Deconvolution is different from Convolution: it's much more complicated and tries to model all stages of the source blurring process. If the blurring process can be accurately modeled, the total blur can be reversed, and the original, sharp image can be reasonably recovered. Deconvolution is very difficult to do correctly, and is used mostly for astronomy and microscope applications. Canon's post tool which recovers edge sharpness, etc., is a form of deconvolution. They can probably do a pretty good job as they have a good idea what the lenses are doing and the entire light path.
 
It's possible to create USM in PPro which is GPU accelerated:


1. Create a copy of the clip and stack above the original.
2. Apply a Gaussian Blur to original. Set to 10 to start.
3. On the upper clip, apply a Brightness & Contrast and set Brightness to 63.7 and Contrast to 50. Set Blend Mode to Difference.


Adjust Gaussian Blur on original to control radius, adjust Opacity on the top video to control amount. It's not quite the same as the built-in USM, however it shows how it's done and that Adobe could include a GPU accelerated version instead of the slow CPU version.


Also try adding an Adjustment Layer (CS6) or Nesting (CS5.x), and applying a (convolution) Sharpen to the Layer or Nested clip.


Tested this technique in CS6 with a 1280x720@60 clip and it brings detail up significantly (and runs in real-time).


Haven't looked at it in depth, but 1280x720@60 ALL-I looks more detailed than IPB in CS6 (don't see that difference between ALL-I and IPB at 1080@24).
 
Last edited:
JCS - Can you use a base 5DMKIII still as an example and show the different methods of sharpening? Probably a lot of work, but I'm interested.
 
You must log in or register to reply here.
Back
Top