Reason I added the sharpen is because the Groom's slightly off focus - not critical, but noticeable. - Arthur.S

duplicate that layer. create a soft mask (or cookie cutter) around the Grooms face & add sharpen to that layer.


OTHER THOUGHTS (random notes, no real organization to them)
note: i'd agree with almost everything Musicvid/Farss have written above.

1) have you tried rendering from a 32-bit project rather than 8-bit to avoid potential moires, etc... your footage is 1440x1080x

2) have you tried using Chroma Blur? (around 0.75 might soften some of the harsher highlights on her face)

3) less dramatic, but you could try dropping Offset 6-10 points (to firm up the shadows) and lower Gamma a touch (5% or so)... this'll give a bit more room in the highlights.

4) colorwise, a small increase in global Saturation would warm up the shots (footage is mostly red w/yellow) and a small amount of magenta in the Highlights may help visually reduce the glare on her forehead...

Just to be clear, this job is done and been well received. Robwood, after all of these probs sharpening, I won't adding a sharpen filter to ANYTHING for the forseeable future.
I've never tried 32bit. How would that help with moire/banding? What's the 12 in 1440X1080X12?

hey, glad to hear all went well

I've never tried 32bit. How would that help with moire/banding? - Arthur.S

the QUICK: projects at higher bit depth reduce the chance of banding being caused within the project.

the LONG&WINDING:
NLE software calculates the result of each filter or opacity setting used; this result is averaged to the nearest RGB value... but when working in 8bit, there are only 256 RGB values available (same as output range)... the lack of extra shades creates the banding as there's little or no wiggle room.... higher bit depths provide that extra calculating room.

hmmm, maybe an easier way to describe this is by math:
a) in 8bit (000-256), to visually describe a range from 100 to 110 you have 11 RGB values
b) in16bit (000-32,678), the same visual range is 12850 to 14135 giving you 1286 RGB values to work with.
c) in 32bit, the number range increases again, etc, etc

as you can imagine creating a natural gradient is easier in 16bit than 8bit.

working at higher bit depths is often done in other disciplines as well, for instance, audio sessions usually work at 32 bit even tho CD is 16bit... graphics are often designed at 16/32bit even tho final distribution may be a JPEG.

in the end you still render to an 8bit file (as usual), but staying at higher bit depth prevents in-project moires/banding from occurring... (tho any banding that was visible in the original footage will still be present)

~

...and the 12 was just me saying this footage isn't even 8bit per channel (256 shades) but only 4bit (16 per channel), meaning more likely an averaging error could occur on the timeline if filters are being applied.

~

arg, this seems confusing even as i re-read it... sorry, i'm not communicating this as clearly as the topic needs, hopefully something in there made sense.

So...
Authored a BD DVD with both the sample files on it Set bitrate to 25Mbps burned to a 2X Verbatim disk and played it on my LG BD player via HDMI to my LG LCD HDTV.

OMG, that is really, really BAD.

I'm inclined to think something has gone awry when DVDA re-encoded the files.
I think Arthur has already tried ripping the files back and checked them but I want to verify this later today myself.

All I can say so far is the artifacts seem related to something happening to the original noise. The bride's flesh is fine, the shadow areas are where the pixels dance.

Bob.

Further results:

Played the BD disk with PowerDVD. The artefacts are definitely there. To see them as clearly as on the Hardware Player > HDTV I had to disable all the de-interlacing in the player.

Ripping the files back from the BD DVD and looking at them with Vegas the artefacts are most certainly there. To see them clearly I looked at them on a 1920 x 1200 monitor. To get a really good look at them I changed the Project Properties from 50i to 50p so I could inspect each field. This revealed that the artefacts change significantly from field to field. It was also pretty clear that most of the components of the artefacts are only single pixel in size.

Viewing the same frames on the internal preview monitor where the frame is being scaled down and fields either dropped or merged they are much harder to see and one could be excused for missing them.

I'd suggest the most fundamental problem has been allowing DVDA to re-encode the video. Second problem was adding any sharpening to what is actually quite noisy video. I still would like to do some more testing to confirm that all the artefacts do originate from noise in the camera original.

Bob.

Final analysis:

Comparing the video from the BD disk as encoded by DVDA with the original source files there is no dramatic difference, the artefacts are in both the source and the DVDA encoded files. To see them well in either requires a full HD monitor and looking at the footage field by field.

The problem really began with the camera which added some nasty compression artefacts to the video. They are not that noticeable without some pixel peeping looking at the footage field by field.

The sharpening by all the FXs tried seems to be working at the field level. This made the artefacts considerably more noticeable by adding single pixel either superblack or superwhite artefacts at the edges.

When burnt to a BD DVD and played back to a HDTV these artefacts then became impossible to ignore as they're quite visually offensive.

Bob.

Thanks, Bob. A little pre-level sharpen usually does no harm, but after rendering his sharpened, unleveled footage in Vegas, then rendering again in in DVDA, (which I didn't know about), then processed by playback decoding could certainly cause the blown pixel tendencies I described earlier).

So, a question, Bob.
If you take his footage, apply a Levels filter (Output Start .008, Output End .922), then render and prepare to BD as you did before, do the sparkles show up? Not exactly definitive, since the original footage was rendered without leveling, but useful information nonetheless.

Bob, thank you so much for taking all of this time and trouble to look into this. One thing though. Yes, I did try DVDA which re-encodes HDV, but only after the problem arose with TAW5 - which doesn't re-encode HDV for BD. I did say that above - honest! :-)

After the original problem arose, I went through the whole job looking for clips I'd sharpened. With XH-A1s you do tend to be fighting the gain a lot indoors, but even a shot of the rings taken outside in good light (that I'd sharpened) had those sparkles. I've also now gone back to the first BD that I produced (about 5 years ago) and checked it through. No problems at all. My work flow hasn't changed, so I'm baffled.

Robwood, yes your post does make sense. :-) So, project settings at 32bit, but render to 8bit? Or have I got the wrong end of the stick??

Musicvid, by "pre level sharpen" are you talking about the order in the FX chain?

"
Yes. In order to do its work correctly, the Levels filter must be applied last, most logically on the Video Output bus.

If I apply a difference mask between Unsharpened and the CK sharpen frames and ramp up the contrast, the "seeds" of the OP's problem are very apparent, including the "pixel clumping" I described above. Do the patterns look familiar?

It looks to me like a basic noise problem, where the camera settings created video that was already over-sharpened, and which had significant noise artifacts. This accentuated whatever sensor noise was already present in the camera's sensor, and all of this was further accentuated by having the bright peaks nailed to 255,255,255. Trying to sharpen this video in post just added to the problem.

In short: a perfect storm.

If it were me, I'd first apply some temporal (not spatial!) noise reduction to the video and then encode that noise-reduced video. I'd skip any post-production sharpening.

Since I don't have any BD production capability, I can't test my theory all the way to the final product, but I did produce a 50i PAL HDV file that resulted from frameserving out of one instance of Vegas (version 10) into an AVISynth noise-reduction script, and then frameserving the results of that script back into a second instance of Vegas. In that second instance of Vegas, I simply dropped the frameserved video on the timeline and then encoded using the standard 50i PAL HDV template. I've uploaded the result to Dropbox, and you can download and see if this "behaves" any better than the original or sharpened videos you've all been using.

Here's a portion of one frame from the original video:



Notice how much noise there is around the bride's temple. There is no way any encoder is going to be able to create decent video when this noise is present.

Here's the same frame after going through my noise reduction:



Here's a link to the noise-reduced video so you can play around with it and see if it looks any better up on the big screen:

Noise Reduced Video

John, In my enhanced version above, I created a 15% composite upper track, which had only very slight gaussian and glow, thus my approach to noise reduction and softening the bride's complexion. On the output, I used [+0.1R, -0.4B] color correction, and finally a levels filter to clamp the whites at 235. I "could" have rolled them off a little, as well.

John, In my enhanced version above, I created a 15% composite upper track, which had only very slight gaussian and glow, thus my approach to noise reduction and softening the bride's complexionInteresting approach to NR. My only objection is that your technique is a spatial noise reduction filter and will tend to soften details. I instead used temporal noise reduction which uses motion estimation to "line up" adjacent frames and then tracks pixel motion between these two adjusted frames. Any pixels which do not follow any coherent movement between these adjusted frames are flagged as noise and reduced/removed. This preserves most of the detail in the original, while still removing the disturbing and distracting noise.

"

Yes, softening and suppressing the lines and pores in the bride's facial highlights was just what I wanted to do with this scene.
In my long-past days of wedding still photography, I would have just smeared Vaseline on the lens . . .
;?)

So, project settings at 32bit, but render to 8bit? - Arthur.S

+1

yep thats the idea... same as setting project properties in After Effects to 16bit tho rendering an 8-bit image-sequence.

Yes, softening the lines and pores in the bride's facial highlights is just what I wanted to do.I was so focused on the noise that I didn't stop to think about the overall aesthetic. I agree that too much detail is not always a good thing, and this may be a good example. Soft lighting, flattering shadows, and a little haze on the lens is a good thing when it comes to bridal photography. Too many skin pores and freckles may not be what one wants.

This reminds me of some of the truly gifted and talented people we used to have on this forum, particularly GMElliott who did (and apparently still does) some of the most beautiful wedding video I've ever seen. Here's his current site:

Cord3Films (GMElliott)

"
"One" in this case being the bride's mother. More often than not, she's the one who decides if the photogs get paid in full.

Yes, I miss GMElliot. Wonder why he disappeared from the forum.

(playback levels)

Unfortunately correcting / grading the video does nothing to get rid of the artefacts.
Needless to say though just looking at it on my reasonably priced, reasonably calibrated monitor it sure does look better after a grade. For my tastes I'd also say the camera original is lightly overexposed. This really shows up in the grooms collar. I'm very averse to blown out white fabric. On the other hand I should mention that in post I can grade up the video from my EX1 quite a lot. Footage from our HC5 and even from a Z5 is another matter. Whenever I suffer from buyer's remorse over how much that EX1 cost remembering this cheers me up :)

Just so we all know what the problem is this image is a cropped and enlarged field grab.



The next field is somewhat different which well explains why these artefacts look so bad on a HDTV.

Bob.

So it's a camera problem?
That one was at the bottom of my guess list.

[I]"So it's a camera problem?"[/I]

I wouldn't blame it all entirely on the camera. It started with what processing and/or compressing the camera did to the image and then what was done with the sharpening added artefacts that are impossible to ignore.

I've struck something similar with the old Sony V1. It seems to have used a lot of dynamic noise reduction. The low light original footage would look quite good but trying to do much with it would show up nasties lurking in the lowlights.

[I]"That one was at the bottom of my guess list."[/I]

The path this topic has taken me down has been quite an eye opener for me too.
As a certifiable pixel peeper I should have seen the problem much sooner.

Bob.

So, to wind this thread up, I'd like to thank you all again for the amount of time you've spent looking into it - especially Bob. It's changed my thinking and will definitely effect my workflow. Maybe not as drastic as I'm sure some of you would like. For instance (and I know I'm gonna get jumped on here :-) ) I don't like the flattening effect that the Broadcast filter/levels have. Of the 2 examples posted by MV, the 'enhanced' version doesn't look HD to me. I like to see the eyes on close ups looking lively and sharp. If folks are paying for Blu-ray, they want to see a clear difference from their DVDs. I remembered a similar discussion I had a while back in another forum: http://www.dvforums.com/forums/latest-behind-scenes-write

I will definitely be restoring the default 'Canon' settings in the cameras, and selectively using levels/broadcast filters. Of the sharpening methods, USM seemed to be the least destructive - and I plain forgot about Neat Video sharpen, even though I'd used it for the speeches, which were done in virtual candlelight. Doooh!! Still very odd (or just lucky) that I've sailed blithely on for 5 years without encountering this before.

You learn a multitude of different skills working as a wedding videographer, as you're almost never (in the UK anyways) dealing with ideal conditions. And a winters day in England in a 14th century house as this one was, is quite typical. Dark room, with the couple in front of a window in direct sunlight.
As always...learn and move on. :-) Thanks again.

I don't suggest using the Broadcast Levels plugin with your camera -- your blacks are already at ~7.5 Setup levels, so the filter would make them look gray and too flat. You should nominally be using .008 Output Start, .922 Output End with your camera in full light. Titles and overlays must also be at [16,235]. If it doesn't look like my histogram above with the preview at Best/Full, then it's not been applied correctly.

It occurred to me while reading your impressions that brides rarely show up HD-ready.
The link to GMElliot's site (above) should give you plenty of ideas and food for thought.
;?)

Yes, applying sharpen post-leveling blows the whites in Vegas.
The plugin chain hierarchy is very important here. That's why the levels go on the output bus in my projects.

VLC decodes with ffmpeg. The choice to use or not use DirectShow (ffdshow) is selectable in the options. Perhaps that has some bearing on your earlier observations WRT Mercalli.