Problems With Still Frame Exports

I'm having a problem creating/exporting freeze frames from a 16:9 DV PAL file in the same sequence settings in FCP 7. I create a freeze frame and export from the viewer via QuickTime Conversion's Still Image, format to TIFF, fps is 25, then set the options to 'no compression.'

But every time I've tried this, there has been a severe loss of quality, the image is rather pixelated. Is it a setting within QuickTime conversion that I'm missing? Or even better, an alternative and better way of exporting stills? Any tips would be helpful, thanks!

If your source is SD DV, the quality will always be bad. Why is your Sequence set to DV PAL? Is this a different project from the other one you were asking about? Because if it's the same project, I thought your source footage was HD 1920x1080?


www.derekmok.com

What derek said. The resolution equates to 72dpi at 720 x 576 pixels, which is very small. You can't do much to get good stills out of DV. If you have a HD version, then go there.

are you deinterlacing



Liezl Wrote:
-------------------------------------------------------
> I'm having a problem creating/exporting freeze
> frames from a 16:9 DV PAL file in the same
> sequence settings in FCP 7. I create a freeze
> frame and export from the viewer via QuickTime
> Conversion's Still Image, format to TIFF, fps is
> 25, then set the options to 'no compression.'
>
> But every time I've tried this, there has been a
> severe loss of quality, the image is rather
> pixelated. Is it a setting within QuickTime
> conversion that I'm missing? Or even better, an
> alternative and better way of exporting stills?
> Any tips would be helpful, thanks!

To Liezl: In the "Export Image Sequence Settings" after selecting TIFF click "Options...". Select "Millions of Colors". Selecting "256 Colors" yields what might be termed "pixelation".

There is one problem with the TIFF export. There seems to be a gamma shift, at least as Leopard displays things. The FCP display of the DVPAL appears lighter than the Preview display of the TIFF.

Quote
Mark@Avolution
are you deinterlacing

This is not relevant since Liezl is comparing the DVPAL image in the FCP viewer with the TIFF image in Preview. FCP displays DVPAL interlaced. Two consecutive fields are shown together as one frame. The exported TIFF has the same interlaced structure.

Quote
Jude Cotter
The resolution equates to 72dpi at 720 x 576 pixels...

720 x 576 pixels would "equate to" 72dpi if the image were displayed or printed at 10" x 8". But it won't be since the image is 16:9, and it is anamorphic. If it is displayed or printed at 10" x 5-5/8", it is 72dpi horizontally but 102dpi vertically.

A 16:9 DVPAL frame has enough pixels for excellent image quality when printed much smaller than 10" x 5-5/8", as it often will be. Small still frame illustrations in books and magazines are examples. It's enough pixels to make a decent picture postcard. It's also enough pixels for good quality poster illustration -- pictures much larger than 10" x 5-5/8" but not viewed from reading distance.

Dennis Couzin
Berlin, Germany

If the source image is HD, it would make a better image, as there is more chroma resolution. Otherwise you can try blurring the u and v channel.

The pixelation could be caused by compression on non legal colors.



www.strypesinpost.com

I really disagree that there is enough resolution in a DV PAL exported frame for a postcard or a poster at any kind of quality. The evidence is obvious. And agree that HD gives much better results.

The thing could be the chroma subsampling, and how QuickTime builds the RGB data, using a chroma smoother in an uncompressed sequence (or export from an unrendered sequence) could help smooth out the chroma blocks, but that's not effective resolution.

Usually for good print quality (posters and stuff), you want 300 ppi, and 720 pixels (with the quarter chroma resolution of DV) makes a decent 0.6 inch wide colored image. If you shot interlaced, it's half rez, since you don't want a fat image, so you make a pretty picture at 0.3 inches wide. For black and white, you get about a 1.2", poster quality image. Of course, everything viewed from a distance looks pretty, which is why there is a Chinese phrase that translates to "from a distance, a beautiful lady, on close inspection, it looks like my mother"



www.strypesinpost.com

> FCP displays DVPAL interlaced.

Only if you view at 100%. when scaled down, it displays only 1 field.



www.strypesinpost.com

Quote
Jude Cotter
I really disagree ...

I just examined three picture postcards sold in museums. Each was printed at 120 dpi for each ink color. I doubt there's advantage to supplying more than double the pixels/inch than the printing dpi. The images are about 4" x 5.5". Then 1320 x 960 is enough pixels to make a museum store quality picture postcard. 720 x 576 is enough pixels to make "a decent picture postcard".

Dennis Couzin
Berlin, Germany

Quote
strypes
Usually for good print quality (posters and stuff), you want 300 ppi,...

Noting your answer is in ppi (pixels per inch) I agree for book illustrations, but not for posters. For movie posters which use frames from the movie -- what we're discussing in here -- to have 300 ppi would require frames on the order of 9000 x 6000 pixels. There have never been such sharp movie frames and the tradition of movie posters has always been with "looks like mother" image quality, except in the fine print.

Quote
strypes
... and 720 pixels (with the quarter chroma resolution of DV) makes a decent 0.6 inch wide colored image.

.
You exaggerate the importance of chroma resolution in image perception. You'd surely find an image rendered 720x576 4:2:0 sharper than the same image rendered 360x288 4:4:4 or 720x288 4:2:2, and yet they have the same chroma resolutions. All chroma subsampling introduces some annoying artifacts, but image sharpness is largely carried by the luminance channel.

Dennis Couzin
Berlin, Germany

It is exaggerated, although with an half inch image, you have full resolution in both chroma and luma channels to build the image.

>All chroma subsampling introduces some annoying artifacts, but image sharpness is largely
>carried by the luminance channel.

To a large degree... Except when you have lots of bright, saturated objects in the frame. DV tends to look pretty blocky due to the low chroma resolution.



www.strypesinpost.com

Movie posters that I can remember are usually graphically enhanced, so they aren't your "let's export to JPG from the movie editor's avid seat and print it poster size" type of job. Also, the images rarely occupy the whole poster.

Where the audience is standing is very important. Even in video- if your audience is going to be seated 100 feet away and they are watching off a 17" screen, it doesn't matter if I shot on the R3D 4K or if I'm playing from a VCD.



www.strypesinpost.com

> Movie posters that I can remember are usually graphically enhanced, so they aren't your
> "let's export to JPG from the movie editor's avid seat and print it poster size" type of job.

I don't do print work, but from what I know, most movie posters don't even use a still from the film as their basis. They use super-high-resolution on-set still photography, and shots from sessions specifically designed for promo stills. Actual film stills might appear on the back of a DVD box -- at the size of a postage stamp. In normal cases, film stills don't often get printed at sizes larger than an 8x11" headshot.


www.derekmok.com

Quote
strypes
DV tends to look pretty blocky due to the low chroma resolution.

It depends more on the decoder than on the 4:2:0 encoder.

Suppose the encoder averages Cb and averages Cr over a 2x2 clump of pixels. DVPAL encoders might not do this, but some 4:2:0 encoders do. Then the chrominance is effectively based on pixels larger than the luminance pixels. Otherwise the 4:2:0 encoder does strict subsampling, as professional 4:2:2 encoding does, with no averaging. (I think averaging has informational advantage over strict subsampling.)

The decoder of the 4:2:0 must decide a Cb and a Cr for each luminance pixel. The brainless way is to put identical Cb and identical Cr at all 4 pixels in the clump, so we get the oversize chrominance pixels in the display. This decoder upres's the chrominance images using what Photoshop calls "nearest neighbor". Perhaps this much pixel oversizing looks "blocky". A smarter way is to upres the 4:2:0 chrominance images using one or another surface-fitting algorithms. Then the Cb and the Cr can change smoothly from luminance pixel to luminance pixel. A smarter yet way is to make the chrominance upres'ing algorithm include both chrominance images and the luminance image in its calculation. I don't know what decoders are in use for 4:2:0, but blockiness is avoidable.

Dennis Couzin
Berlin, Germany

>I don't know what decoders are in use for 4:2:0, but blockiness is avoidable.

There will be some artifacting. It will be either blocks or blurriness, depending on what is used for the decoding of the signal. This is caused by undersampling the chroma channel. A better decoder (or a good chroma scaling algorithm) can do this and that there is a properly sampled luma channel, which can help decode the chroma.

Interpolated resolution is not actual resolution. Dv has quarter chroma resolution, so it looks good for some images and are horrible for some, and is also pretty bad for image processing or print.Of course, if the OP's image is shot interlaced, there will be a lot of interpolation, effectively 1/8th chroma resolution and half luma resolution.

My numbers previously is exaggerated (half inch would be the actual physical size of an image at 300ppi with no interpolation. On a suitable image (not highly saturated sunsets or where the image fidelity is heavily dependent on color), a decent 2-3 inch image should be possible on an interlaced image (150 pixels per inch).



www.strypesinpost.com

Quote
derekmok
...most movie posters don't even use a still from the film as their basis. They use super-high-resolution on-set still photography, and shots from sessions specifically designed for promo stills.

There are the annoying movie posters unrelated to stills from the film. They're to lure people into the theaters but not to keep as souvenirs of a movie enjoyed. But I must agree that even the posters seemingly based on stills from the film can be otherwise. I just now checked the souvenir poster from "Taxi Driver" versus the film and it's not quite the shot in the film.

Dennis Couzin
Berlin, Germany

> They're to lure people into the theaters but not to keep as souvenirs of a movie enjoyed.

That's the function of a movie poster. Advertising.


www.derekmok.com

Quote
strypes
Interpolated resolution is not actual resolution. Dv has quarter chroma resolution, so it looks good for some images and are horrible for some, and is also pretty bad for image processing or print.Of course, if the OP's image is shot interlaced, there will be a lot of interpolation, effectively 1/8th chroma resolution and half luma resolution.

To call 4:2:0 "quarter chroma resolution" begs misunderstanding, since the word "resolution" standardly refers to linear resolution. For example, you mentioned pixels-per-inch, not pixels-per-square-inch. 4:2:0 is best described as half chroma resolution horizontally and half chroma resolution vertically.

4:2:2 is half chroma resolution horizontally and full chroma resolution vertically. I find it odd that with their difference just in vertical resolution, 4:2:2 is generally professionally acceptable while 4:2:0 is utterly scorned.

Interlacing doesn't imply interpolation when displayed on CRTs for which interlacing was intended. There's no filling in of missing lines at either the retinal or brain levels of visual processing. What goes on isn't interpolation, and you shouldn't halve the vertical resolution in case of interlacing.

Interlaced video displayed on other displays is typically deinterlaced. Deinterlacing algorithms that look at fields before and after the field being deinterlaced don't result in a halving of the vertical resolution. Consider, for example, intelligent deinterlacing of a static or nearly static scene. Then there's full vertical resolution.

Interpolated resolution isn't actual resolution, but neither is it pure figment. Interpolation is based on assumptions about the world: if the real scene put this much light into this pixel, this much into the next pixel, etc., then how much light would the real scene have put into half-pixels on the same array? The mathematical assumptions about the world are usually approximately correct, and rarely horribly off. When subsampled chroma are interpolated, as must be done for display, the actual resolution in the luma can make the interpolated chroma resolution actual resolution. For example, in real scenes the luminance "edges" generally coincide with chrominance "edges". (Chromatic aberration in lenses may blow this.)

Dennis Couzin
Berlin, Germany

>Interlacing doesn't imply interpolation when displayed on CRTs for which interlacing was
>intended

In this case, we are talking about print, where the frame will be closely scrutinized. Yes, de-interlace, as it will look better.

Interpolation is approximately correct. I guess that is the best we can say about interpolation. There will be errors. Have you compared text that is de-interlaced to text that is generated as progressive? Or footage after an up conversion? It looks worse if you are up converting from an interlaced image, than if you were up converting from a progressive image. There is more spatial interpolation done if the image is interlaced. My guess is that they deinterlace each field, up convert it, then re interlace the image. Good algorithms improve the quality of the interpolation.

In this case, I am not sure how many declares there are that use the luma channel to re-interpolate the chroma channel. My guess is that they don't.

Interpolation can be relatively accurate. The red codec is relatively low noise, and it uses interpolation to debayer the image. Dv is kind of ancient. And many graphics people I know of blur the chroma channel slightly to get a smoother key.



www.strypesinpost.com

to strypes:
Good quality deinterlacing involves both spatial and temporal interpolation. An empty line in a field is surrounded (spatially) by nonempty lines above and below and surrounded (temporally) by nonempty lines before and after.

Liezl worked with DVPAL: interlaced. Liezl's first mistake was to "create a freeze frame" from which to export a still image. A freeze frame created in FCP using Modify > Make Freeze Frame discards the bottom field. If he had instead exported straight from a frame in the video the still would contain both fields. In neither case, however, does the exported still get information from the frames before and after the original frame. So it is too late to do a proper deinterlacing of the still image. Photoshop can only do a blurring or pixel-doubling deinterlace on the two-field export.

To get the best still exports from interlaced video one should first use spatio-temporal deinterlacing on the footage and then export a progressive frame.

Dennis Couzin
Berlin, Germany

>A freeze frame created in FCP using Modify > Make Freeze Frame discards the bottom field.

Nope. It keeps both fields in the image.



www.strypesinpost.com

Quote
strypes
Nope. It keeps both fields in the image.

I double checked. There is only the upper field in the freeze made by Modify > Make Freeze Frame.

Did I do something wrong in FCP to get that? Then it's something Liezl could also have done wrong.

The first point is that a freeze frame which discards one field doesn't yield a good still. The second point is that a freeze frame which keeps both fields also doesn't make a good still because there is no good way to deinterlace from just two fields. (This assumes there is some motion in the frame.)

Dennis Couzin
Berlin, Germany

Make sure display is set to 100%, or you won't see the fields.

Also, make sure easy setup is set correctly. I was on 24psf and make freeze frame became progressive. But yea, I have to deinterlace freeze frames whenever I use them in a project or I end up with stuttery freeze frames.



www.strypesinpost.com

definitely freeze frames & frame exports carry both fields.

as Strypes says, you actually need to de-interlace freeze frames made in FCP
that was one of the hugely Frequently Asked Questions for about the first 5 years of FCP.

and frame exports brought into Photoshop need de-interlacing

interesting idea about accessing adjacent frames for more video info / less video noise
i know of some filters that do this (Joes Filters has something along these lines)
i didn't know it was part of De-Interlacing strategy.

FWIW, the Nattress de-interlce is the best for FCP IMO.
not sure if that accesses adjacent frames.
its very clean, which would suggest to me that it doesn't,
(also there are no controls for that function)



nick

I think the deinterlacer in Compressor is pretty clean once you set it to "best". I'm not sure if there are any good chroma scalers for DV in the market. Most of them smoothen the chroma channel (blur), so I guess you would run that after you deinterlace.



www.strypesinpost.com

strypes & Nick: I see how it happened. I double clicked on the frame in the canvas. It then appeared in the viewer, still interlaced. Then Modify>Make Freeze Frame. The lower field disappeared in the viewer and is missing from the created freeze frame. In order for the created freeze frame to get both fields, the canvas, not the viewer, must be selected when applying Modify>Make Freeze Frame.

This is not how the User Manual says it should work, but I find it so. The sequence settings are standard DV-PAL. See what your systems do.

Googling <<spatio temporal deinterlacing>> shows that after more than 20 years it's still a lively topic.

Dennis Couzin
Berlin, Germany