Results of still photo flicker tests

johnmeyer wrote on 5/20/2005, 5:53 PM

Several times a week on this forum, people ask how to reduce flicker in still photos that are moved using pan/crop. I have a huge project with over 1,000 still photos, and I wanted to make sure, before I rendered, that I was doing everything correctly. I found the worst photo of the bunch -- worst in terms of the artifacts created when using pan/crop to move it -- and then tried all sorts of things to eliminate the problems.

This is not an exhaustive test. I'm sure the results will depend on the source material, file format, render type, and many other variables. However, you may find the results useful.

I took my test photo and created some pan/crop moves (zoom, diagonal move, rotation). I then duplicated this and then tried various things, one at a time. Here are the things that made the most difference, starting with the one that made the biggest difference. I rendered to NTSC, created a DVD using DVDA, and then played through my DVD player onto a 45" television monitor.

1. Reduce resolution before importing into Vegas. This helped the most. My original file was a 1536 x 2048 JPEG image. I used "Match Output Aspect Ratio" in the pan/crop which gave me 1536x1126. If I hadn’t zoomed, then I could have reduced the resolution, using PhotoShop, by 50% and still have had enough pixels. However, since my pan/crop movements included a zoom of about 30%, I needed an image with about 30% more resolution than the 720x480 of my NTSC final output. By reducing the image to 61% of its original size (in Photoshop), I ended up with a 937 x 1249 image. This left just enough resolution to zoom in 30% and not end up with an image that had fewer pixels in either direction than 720x480.

The flicker, staircasing, and other artifacts around hard transitions from light to dark, were dramatically reduced using this technique. I didn't see any noticeable new artifacts, nor did the image seem to be degraded. One other benefit is that the lower-res files render faster.

2. Reduce Interlace Flicker. I was surprised to see that this helped quite a bit. It is definitely worth doing. I didn't see any noticeable new artifacts, nor did the image seem to be degraded. There are several scripts bouncing around various sites that automate the process of applying this attribute (I am at the moment too lazy to find the links — sorry).

3. Gaussian Blur. I applied the Gaussian Blur fX and used the smallest possible setting (0.001). This definitely helped but, even using this small amount, it slightly reduced the "edge" of the picture. I wouldn't call it soft or out of focus, but it just wasn't quite as snappy as before. I have seen a number of different suggestions about using other types of blurring, and I didn't have time to experiment with these. Perhaps someone else would like to experiment. I was less intrigued with this approach because it is obvious that the picture will be degraded. Also, rendering times using Gaussian Blur were quite a bit longer.

4. Best Rendering. When you choose "Render As" and then click the "Custom" button, the first option you see on the Project tab is "Video Rendering Quality." This is normally set to "Good." The Vegas help system has this to say about this setting: "If you're using high-resolution stills (or video) that will be scaled down to the final output size, choosing Best can prevent artifacts." Based on this, and on posts in this forum, I have been recommending to people to use the Best setting when rendering a project that contains lots of high-resolution still images. While the results were better than with good, the differences were quite subtle compared to 1-3 above, and the rendering time was MUCH longer (1:48 for good and 3:23 for best).

5. Supersampling. I have read several posts that claim the Video Bus supersampling envelope can help. I applied it to the original picture, but was unable to detect any noticeable improvement. At least in this one test case, Supersampling was a waste of time.

Conclusion For 95% of "problem" images, reducing resolution down to 720x480, multiplied by the maximum zoom you plan to use, gives you the most bang for the buck, although it requires you to spend time with each image in PhotoShop (or your favorite editor which, for me, is PhotoImpact). If you combine this with Reduce Interlace flicker, even a problem image like the one I used can be made almost artifact free. Taking the next step and using resolution reduction and reduce interlace flicker along with Best rendering gets you very close to no artifacts at all. If you absolutely must kill every last bit of flicker, then go ahead and apply some form of blur (as already noted, I used the Gaussian Blur at the 0.001 setting). All four settings together will pretty much kill flicker in any image.

One related note: As Ed Troxel noted in this thread (), you should use Pan/Crop to generate motion with images and not track motion because track motion operates at video resolution, after the still has already been down-sampled. You can easily verify this for yourself by putting the same image on two tracks, one above the other. On the top track, zoom way in on some detail in your picture using the pan/crop event setting. On the track below, use track motion. The track motion zoom will be totally blurry compared to pan/crop. For video, this problem doesn’t exist at all because video is already, by definition, at video resolution. This issue only applies when the source material is high resolution or generated by Vegas itself.

[Edit] After many people replied, it was pointed out that you only need to set the vertical amount of Gaussian blur to 0.001, and leave the horizontal setting at zero. This should provide most of the flicker reduction, without as much softening of the image.

Comments

Chienworks wrote on 5/20/2005, 5:57 PM

#5, as you suspect, supersampling has no effect whatsoever in this context. Supersampling only effects temporal resolution (frame rate) and does absolutely zilch with spacial resolution (frame size).

Michael L wrote on 5/20/2005, 6:57 PM

Thank you for the great overview. I have seen each of the suggestions at various times but have never seen a comparison as thorough as yours.

It will be a reference.

craftech wrote on 5/20/2005, 7:09 PM

John,

Thanks as always for your tireless dedication to helping the people on this forum.

John

farss wrote on 5/20/2005, 7:25 PM

John,
in my case the output format was 16:9 PAL DVD with no zooming or motion of the stills. My goal was to get an image that was every bit as good as if it was shot with a 16:9 broadcast camera, in fact I might have been even able to do better, after all I was starting from 4.4.4! As there was very little money in the job I needed a quick to do solution, render time wasn't an issue, the Monster could be left running at zero cost for days if need be.
The client had supplied an EDL and the stills on CD, fiddling with the images in PS would have been very time consumming, I guess I should learn how to automate PS!. But the other issue I had was some of the stills the client had already cropped heavily, going down the downscaling in PS route would mean having to check every image beforehand. Also some of the images were soft and didn't need any treatment. What I needed was a way to work that didn't knock anything down anymore than it needed and still gave an excellent result, the final product looks like it's going to be sold world wide.

Using GB in a SD project does seem to reduce the res more than I liked, switching the project to HD seems to change the way the GB is calculated, maybe I'm wrong as I've never had a conclusive answer to this one. Certainly the best shot stills come up looking as good as anything shot with a very expensive video camera, as they should.

Bob.

Cheesehole wrote on 5/20/2005, 7:44 PM

re: 3. Gaussian Blur

Also check out the Median Filter. It's slow, but it gives some blur options that seem to preserve more detail overall - I think anyway. I also use it to reduce pixel pumping in MPEG-2's when the photo is frozen.

I haven't made a photo montage in a long time - thanks for the great summary.

I wish they'd change it to "Frame Supersampling" or something :-Z

johnmeyer wrote on 5/20/2005, 8:18 PM

Also check out the Median Filter.

Thanks. That is something I didn't have time to experiment with.

And, farss, I sympathize with the problem of not wanting to down-res dozens (or hundreds) of pictures, especially if each one is a different resolution and each is going to be zoomed by a different amount. This means that every picture will need a different down-sample amount in PhotoShop. The "reduce interlace" combined with some sort of blur is definitely the most time-effective solution, especially if one of these other blurs has less of a negative effect on the sharpness.

Laurence wrote on 5/20/2005, 9:20 PM

Just to confirm, with the gaussian blur test you applied the smallest amount of blur vertically and none at all horizontally right? That is the setting that works the best.

TeetimeNC wrote on 5/21/2005, 4:27 AM

Thanks for posting this John. I just went through a similar experiment with the same findings, and I am sitting here watching the 12 hour rendering. In my video I am panning across a scan of a road map - very challenging to get out all the artifacts. I have the project set to Best, but wondered after I started the render if it would have worked to render just this map to the timeline using Best, then switch to Good to render the whole project. Also, I wish I had tried GB vertical only - I am using .001 vertical and horizontal.

-jerry

JJKizak wrote on 5/21/2005, 5:35 AM

I have placed huge jpg slides on the Vegas timeline with very mixed results. 500k pictures seem to be for the most part trouble free but when I hit 1k or 1.5k sometimes they are fine, sometimes not. It seems as though there is some kind of resonance in some pictures as far as size and pixel count that when you zoom sometimes no flicker and sometimes there is tremendous flicker. Mostly in the horizontal plane.
I see this a lot on TV commercials even with bigtime Ford, GM, etc.
and it occurrs only during panning or zooming. Sometimes resizing will work and sometimes not. It's almost like trying to scan the dots on the inkjet printed pictures.

JJK

BillyBoy wrote on 5/21/2005, 7:31 AM

I've commented on this many times in the past and attempted to summarize both the science and the human element but due to space limitations of the forum format getting to all the details if you interested is better served by visiting web sites that cover it in more detail.

So the short and long answer. The short answer is flicker happens because of how television was originally designed which became various "standards' depending on where you live and also on how your eyes and brain process what you see.

You can get a better understanding of the "why" at many web site, some that I offered before or that you can find on your own. The one below takes a somewhat different approach. I at least found it interesting reading. It also backs up what I've said many times about CRT type TV's being a poor choice and inferior to Plasma or LCD due to the advantage of sample-and-hold characteristics.

http://www.microsoft.com/whdc/archive/TempRate.mspx

johnmeyer wrote on 5/21/2005, 8:46 AM

Just to confirm, with the gaussian blur test you applied the smallest amount of blur vertically and none at all horizontally right?

Hmmm .... No, I did 0.001 for both vertical and horizontal. As I think about it, I can see exactly why horizontal might not be needed. I'll try vertical only next time. Perhaps the qulaity reduction won't be as severe. As farss already pointed out, the time required to down-res the pictures is a pretty severe limitation, even though it produces the best results. If reduce interlace flicker combined with blur and best rendering can get results that are almost as good, then that might be a better way to go for many people.

Thanks for the tip.

Randy Brown wrote on 5/21/2005, 9:00 AM

Nice, informative thread worth bookmarking....thanks guys.
Randy

fultro wrote on 5/21/2005, 9:00 AM

I have found that horizontal is not needed and using only .001 Vertical reduces most flicker problems and doesn't soften most images too much whereas adding the horizontal into the mix definitely does. For many things (not all) I have been able to get away with just this vertical blur and only downrezing the occassional problem photos. On the other hand I have literally never seen any diff with the Reduce Interlace Flicker switch.. I now want to try this Median Filter idea ....
Thanks for your work John .. it is nice to compare notes with others when experimenting and (for me ) shooting in the dark....

Laurence wrote on 5/21/2005, 9:10 AM

Oh! I almost forgot something incredibly important!. On a high resolution scan, if it was a scan of anything other than a photolab print, you need to zoom way in and get rid of the dot pattern from whatever it was printed with. You do this by zooming in to where you can see the actual pixels and applying just enough blur to make the dot pattern disappear.

If you don't do this you get all sorts of artifacts on any motion as the pan and zoom goes in a binary fashion from blank space to separate colored dots in each primary color. A high resolution scan of something like a map will just crawl with artifacts on motion if you don't do this.

I do this when I scan so I've always thought of this as more of a Photoshop than Vegas step. Some scanning software has a function like this built into the scanning, but I usually turn it off because a manually applied blur is more precise. If you do this step, you get smooth photo animation regardless of how big the scan is. I prefer it to downsizing the scan (which works because it is clumsily doing the same thing).

After applying the fine gaussian blur, there will still be a high amount of detail which will interfere with the interlace lines on regular 60i format. This is where the .001 vertical blur removes the remaining problems. Another option is just to render the project 24p which will get rid of the problem entirely when a DVD player and TV is set to progressive scan, and not be all that bad even when it isn't.

Scanned photolab prints don't have this problem because picture is not made up of tiny dots like a computer or printing press printout is.

johnmeyer wrote on 5/21/2005, 9:29 AM

On the other hand I have literally never seen any diff with the Reduce Interlace Flicker switch..

Prior to this experiment, that was my impression as well, and I even posted something recently that was almost identical to your comment. Perhaps the effect of this control depends on the reoslution or nature of the image. In this test, however, it made a VERY noticeable difference.

BillyBoy wrote on 5/21/2005, 9:37 AM

In real estate its:

Location
Location
Location

With flicker and artifacts one main thing determines how successful you will or won't be getting rid of it or your success in reducing it.

The source file
The source file
The source file

riredale wrote on 5/21/2005, 9:49 AM

One other thing I noticed: if you are doing a significant zoom-in or out on a still, you need to make sure you have enough resolution for the extreme zoom-in situation (as you've mentioned). The problem is that for substantial zoom ratios, that means you will have an excess amount of resolution on the full zoom-out, with the resulting artifacts (which I believe is officially termed "spatial aliasing"). The necessary solution is to keyframe Gaussian blur so as you pull out you are also softening the image. I've had to do something like this all the time with many of my PIP video shots where the video starts out full-screen but then shrinks down to a size maybe 20% of original size in order to act as a part of a moving menu for DVD links. If you don't gradually keyframe in Gaussian Blur you will see all kinds of aliasing in the small video.

Laurence wrote on 5/21/2005, 10:21 AM

Just apply enough gaussian blur to the source scan to get rid of the scanned dot pattern, and almost all the other problems disappear. The only thing left is detail that is so fine, that it appears on one scan line, but not the interlaced scan line above or below it. A .001vertical gaussian fixes that. Again if the output is 24p you can skip this step. Any other way to fix this problem is not going to be as good.

fultro wrote on 5/21/2005, 11:41 AM

Laurence - I must ry this - I wish I had known a month ago when I had to scan a bunch of half -tones ...

and John - you say that by doing step one, step two becomes way more noticeable - not to be nit-picky but if step one as you say takes care of almost all problems I wonder how step two all of a sudden becomes so significant - I guess I'll just have to try it the next time I 'm working this way....

GlennChan wrote on 5/21/2005, 12:03 PM

Reduce interlace flicker: By default, this does nothing.
Go into file --> project properties. Set the de-interlace mode to blend fields.

An alternative to reduce interlace flicker (Vegas term for de-interlace) is mike crash's de-interlace filter. http://mikecrash.wz.cz/
It maintains more vertical resolution.

2- Another option is to try the unsharp mask filter. You can actually enter in negative values, and values greater than 1. Try:
amount =-0.500 (for negative values, this is the strongest. -0.4, -0.3, etc are less strong)
radius = 0.003
threshold = 0.040

This will blur out high contrast lines, both vertical and horizontal. Blurring the vertical detail will get rid of interlace flicker. Blurring the horizontal detail will help reduce chroma crawl (use RCA/composite connections to see chroma crawl).

I don't work with pan and scan projects so I don't know how well that works. But it seems to me to be a really good way to deal with your problems.

3- When scanning certain types of printed images, a (de-)moire filter in Photoshop is supposed to deal with offset printing. You can run batch actions in Photoshop to automate the process.

fultro wrote on 5/21/2005, 2:29 PM

"An alternative to reduce interlace flicker (Vegas term for de-interlace....."

Is this true ? because correct me if I am wrong but it still has to be re-interlaced for TV - so this doesn't make sense to me

kchaz wrote on 7/19/2005, 3:57 PM

excellent thread! thanks for info!

couple other things to consider, and has helped in my photo montages, is still digital images have square pixles, whereas TVs use rectangular pixels....so by changing the still image pixels rectangular the image become better suited to view from TV, ...in photoshop, i always do two things (as a batch process) to all stills that i am going to use in video...

1. i make sure i apply NTSC color from photoshop (filters>video>ntsc colors), to make sure my image colors will be right for TV viewing

2. go into pixel aspect ratio and change it to DV, NTSC (whatever flavor...this will chage your pixels to rectangular, which looks a bit strange on computer monitor, but perfect on TV or external viewer.....

JJKizak wrote on 7/19/2005, 4:44 PM

I have been trying to associate the physical dimensions of the picture---2200 x 2200, 720 x 480, 1980 x 1280, 800 x 800, etc. as applied on the timeline with match aspect ratio applied and have not finalized anything yet but strongly suspect that certain sizes create the squigglies. Slower panning and irregular velocity in zooming at the same time can reduce the affect.

JJK

lg777 wrote on 12/5/2005, 6:58 AM

Just curious though, this seems like a bit of effort to get photo stills to display properly but there are a handful of slideshow programs that do this without any prep...such as Proshow Gold and MOTV. I've used these programs a lot and find good and consistent results. I also incorporate video clips as well but these are not good video programs at all.

Just can't understand why it's not the same for a package like Vegas to do the same. Handling of still photos that is. With a show mixed with video and about 50 stills, this gets a bit tedious.

Love Vegas for video though. Just got Vegas 6...was using Vegas Studio 4.0