Focus, unblur, sharpen, adjust colours

[SearchQuality]

Hi folks. Recently I had to shoot a clip under somewhat difficult conditions. There was not enough light, I could not approach the target, there were objects in front of the camera, which spoiled the focus etc.. Now I am back at my desk trying to do whatever is necessary to increase the quality of the MPEG file. Focus needs to be improved, unblurring, sharpening is necessary and colours need to be adjusted. I am using Vegas Movie Studio 10 Platinum edition. I am also quite familiar with VirtualDub and Avisynth. I can make use of GIMP as a replacement to Photoshop should this be necessary. Any help is highly welcome. Thanks in advance.

[poisondeathray]

Unless you post an original clip or give a better description, you're probably not going to get very specific help

Otherwise you can try sharpen, levels , color correction filters - but you already know this

[SearchQuality]

Of course. Here the file is attached.

[sanlyn]

I've always found that using fully automatic cameras in anything other than average daylight lighting is not your friend. Rather, it's your worst enemy. What I get from viewing this video and trying a few things served to illustrate some basic fracts. The human eye and brain can accept and adjust to a vast range of contrast, mixed lighting, and color balance situations. But video, whether film or DV, doesn't have this ability. If you can used your camera to set exposure and lighting only for the center of interest -- which I assume would have been the activities everyone was watching -- the audience being dark wouldn't matter. You see Hollywood shooting scenes like these that make everything clearly visible and "within spec". What you don't see is all the lighting and reflectors behind the cameras to make it look that way.

In these shots your auto camera did exactly what it was designed to do: it determined a middle-gray area based on the contrast range in front of the lens, and it let other elements (very darks and very brights) fall where they may. The contrast range between audience and center stage was wide to the point where there is no detail in the dark audience nor any detail in the bright faces. Those elements are clipped. By "clipped", one means that video data that would fully described those extremes is not there. That is, there's no data in those areas. You can use various techniques to brighten the darks, but there will be no detail and not much color -- they will simply be brighter. You can darken the bright faces, but there is no data in the brightest highlights.

It didn't help that no filter or camera setting was used to correct color balance in this scene. I believe some cameras have settings for various lighting situations. This camera was apparently set for daylight, which is a much higher color temperature than the lighting in this scene. The audience is orange because the light falling on them had very little blue in it.

There are varying mask techniques that can be used in more sophisticated sopftware, but even there you are dealing with this: a frame from your original, no processing. . .

[Attachment 13244 - Click to enlarge]

I used some ColorYUV settings to raise detail in the darks and lower the brights. Then a gradation curve filter to tame some of the red. This works to a limited extent. I think you can, see however, that lowering brights and raising darks did little to bring any detail into those areas. You can use sharpening, but mostly you'll get grain and just sharpen noise. You can't create detail where none exists. The lack of detail isn't a matter of focus; the murky shadows and faceless faces are symptomatic of severe clipping. There's not even enough blue bdfata in the orangy audience to keep highlights from looking discolored, and a few of the "people" details are turning green.

[Attachment 13245 - Click to enlarge]

Using SmoothLevels() to keep luma and chroma within a valid video range and some ColorYUV to tweak things even more, the audience is abandoned and left-as in favor of taming the bright faces and raising midtones. The brightest parts of this image have been lowered to RGB220 or below to try to retrieve some original highlight detail. It just isn't there.

[Attachment 13246 - Click to enlarge]

[SearchQuality]

Thanks for this analysis Sanlyn. To summarise your post: the detail data is not there and something that does not exist can not be revealed. I should at least have set the focus and white balance parameters manually. Can you tell me which tools and filters you would have used had the data been there?

[sanlyn]

Most members here would do as I did: convert the mpg to AVI with lossless Lagarith compression. I used DGIndex to make that conversion. This AVI was in its original YV12 colorspace. In Avisynth I used ColorYUV and SmoothLevels (part of the SmoothAdjust plugin package). The SmoothLevels() and ColorYUV statements I used to produce the bottom image of the 3 posted above were:

SmoothLevels(tvrange=true)
ColorYUV(gain_y=-18,cont_y=-40,off_y=-17,gamma_y=15,cont_v=-70)

Those were were derived less for color "correction" than to illustrate a point. To view the results of the efforts I viewed the output in VirtualDub using Trevlac's ColorTools RGB histogram/vectorscope/waveform plugin and a gradation curves color filter, the latter used in the middle image, above. The RGB tools are similar to those found in Photoshop Pro and After Effects. Avisynth has useful histograms and other 'scopes, too. But I like viewing all 3 colors + luma in RGB, even though I don't always go to RGB for correction tools. With YV12 or YUY2 sources having bad levels and chroma problems, it's almost essential to correct in YUV first, before moving on to denoising or working in other colorspaces.

If the video had been well exposed but, say, too grainy or otherwise noisy, it would be impossible to recommend any one or two filters. Most have a particular purpose, others are multi-faceted but can be tweaked/modified to your heart's content. For HD video of this sort, I would stay away from strong sharpeners; most people tend to overdo sharpening anyway and end up with rather odd-looking results. Video shot under conditions like those above usually have noisy shadow areas, so I'd try MCTemporalDenoise at low settings but would turn on its debanding feature, Or, a touch of Avisynth's TemporalSoften might just be enough. Sometimes a very light touch of NeatVideo (RGB only, and not free) cures many ills, but it's a product that most owners abuse, with abusive results.

To get a feel for many of Avisynth's more popular plugins, you might browse this AMV page, with good samples and demos: http://www.animemusicvideos.org/guides/avtech31/post-qual.html . It's an anime website, but the principles and plugins can be used with any video.

You should see the wrecked video I made at my niece's graduation a few years back. I had always been a manual-set Nikon guy, but the wife had this new auto cam and she talked me into shootin' away, helter-skelter. What a mess. But, then, no -- maybe you shouldn't see that video.

[SearchQuality]

Thanks for the very useful advice. I have a Sony Handycam HDD DCR-SR37 bought in 2009. I used the latter to shoot this graduation ceremony. It does everything automatically. But the result as you see is a disaster. A number of parameters can be set manually. Spot mtr/focus, spot meter, spot focus, exposure, focus, white balance and scene selection with twilight, candle, sunrise, sunset, fireworks, landscape, portrait, spotlight, sports, beach, snow are available. If you had to shoot the same scene under same conditions which parameters would you set manually and/or which scene selection would you prefer? Thanks.

[sanlyn]

focus: If you can manually lock focus on the central action (rather than near objects), that would help. Let's face it, under low light your lens aperture will have to be wide open. Thus, depth of field is limited with wide apertures, so trying to get everything in focus from near to medium to distant isn't possible. From where you seem to be seated in this video, a medium distance would do it. If I recall, you can use autofocus or spot focus to determine an area of best focus, then turn focus lock ON to maintain that setting. Once you do this a few times it gets faster and easier, so not difficult to change when you must.

I'm surprised there's no tungsten or indoor light setting. "Spotlight" is probably closest to what you need. Candlelight is too red, daylight is too blue. But remember you're in a mixed-light environment, so don't expect perfection in every area.

Spot metering is difficult for most people to grasp. The meter reads the light in whatever spot (usually center-spot) you aim for. The reading will give the correct exposure for middle-gray objects: that is, anything that's suppose to be properly exposed and that is actually of middle-gray luminance would be correctly exposed. The spot doesn't have to be gray (the meter reads light values, not color values). What many photogs use for objects with approximately middle light value are shadows on skin tones, most grass or shrubbery not in direct sunlight, the shadow side of brightly colored objects, the light-grayish shadow areas of snow in winter scenes, and so on. If you measure the brightest part of a picture (such as the directly-lit side of the face of a person facing the sun, the brightest part of a sunlit sidewalk, or the brightest face in a spotlighted scene), the meter decides how to make those bright objects a middle value -- exactly what you don't want, as the meter is telling you what settings will make bright white look middle gray! So, using spot meters requires some experience. If you took a reading off the bright faces, you'd need an additional 2 to 2.5 f-stops wider lens opening to keep the face bright without washing out. Or, if you're using the "averaging meter", you could have stood up in place and taken a reading off the main activity area, which had bright, middle, and not-too-dark objects mixed; if you're careful not to let large dark objects or areas dominate the image, that average should get you in the neighborhood of what you want.

I used to carry a Kodak gray card around with me. This is a 5x7 or so cardboard with middle gray on one side and 90% (almost white) on the other. Too bad plastic film cans aren't around today: those who were paying attention noticed that the snap-off top of those plastic film containers were middle gray! I also had since 1980 a pocket-size (actually a little larger, but no matter) Kodak Pocket Photo Guide with 16 pages of the most useful data and exposure cards you ever saw. Those were sold in small specialty shops that catered to advanced amateurs or better, not in big-box electronics stores. You can use still-photo guides as well, since the principles of exposure are the same for video and photos alike. You might wonder why a scene with lots of color is measured with a gray object. It's because middle gray contains all three colors in equal portion. The measuring area doesn't have to fill the viewfinder; bringing it close enough to dominate the image, even if the gray is out of focus, still works. Your light meter doesn't care about focus.

Despite all the measuring and calculating, nothing beats some trial and error. When you have time, shoot some video forn the heck of it and examine the results. It doesn't take long to develop an eye for exposure problems.

[SearchQuality]

Thanks a lot for your contribution Sanlyn. I shall keep your posts as reference pages in my video guidance archive.