Does anyone have any solutions to contribute? I am sure the keeners around here have seen the s-pline pro footage...wow

but, even if I was going to purchase it, the program needs frames and thats just unrealistic even for 5mins of video...unless we can find some way to automate the process

i guess it would be a good time to ask anyone found any alternatives as well?

thanks for your time,
Rob

Hi Rob,

S-Spline has an option to batch process any number of frames. Essentially, you export your timeline as an image sequence compatible with s-spline, then run s-spline against that image sequence into a new directory.

Re-import the newly uprezzed images into your NLE as an image sequence and render a new movie file.

If you have the drive space, i would avoid using any compression on both passes.

I tested the process on a demo copy at work...I can't do batch processing using the demo, but a frame took about 7 seconds to rez up to HD resolution. Including opening and saving the image, take that up to maybe 11 seconds per frame. Might be faster.

So 11 secs per frame x 30 fps x 60 seconds x 5 mins is, what, 5.5 hours? Did I do that math correctly? So run the batches overnight.

S-Spline doesn't batch process, but S-Spline Pro does.

They're also going to be coming out with a video processing program that will likely do exactly what you want.

With TARGA and TIFF files you can use non-destructive compression
if both sides support it. Smaller files without quality loss. Might
help if uncompressed is a bit too much.

Barry, you're right, that distinction would make a big difference when someone went to buy the software.

Rob, I couldn't find any information on what their intention for video processing would be. Did they give you a timeline?

Also, the following link was provided in an uprez post in the MX list:
http://www.aequantum.com/kiddo.htm

I find the bulk of uprezzed images to increase dramatically in contrast and video sharpness, as in the sharpness that we all try to turn down on their cameras.

Has anyone found a method for smoothing this abberation without running a destructive gaussian blur on the image?

Another possibility is Fred Miranda's Stair Interpolation Pro, which runs $20. I haven't tried it, but it might be worth a look.
http://www.fredmiranda.com/shopping/SIpro

that other one looked interesting too...i was hoping there would be some sort of video-friendly program for this sort of thing though

what about virtualdub? any plugins like this? there are tons....i myself havn't come across any...but even still, is this sort of thing tough to program? probably :S

thanks again
Rob

I tried it out, running it with the B-Spline option. (The S-Spline gave to much sharpness) It works better on some particular images.

I uppresed the video to 1880 and run it back down to PAL. (720)
and I could not see any diffrence between the native and the upprezzed material downconverted again.

I thought perhaps I could give my video a little more fake detail by doing this, but there was no diffrence..

Good Idea if it would work though...

Andreas: I have ran some tests on frames in Photoshop, using the technique of small increment bicubic interpolation. While the uprezzed frame looks good, after sizing it back down to 720x480, I can see no real difference between it and the original.

However, it does have advantages. Run your grain filters or sharpeners (assuming the footage was captured without electronic sharpening to begin with) on the big footage and then drop it back down in size. This would look better than just running the filters on the 720x480 frame.

The guys that transfer DV to film are bound to be using some sort of interpolation process similar to what we're talking about.

Anyone know?


,Frank

Here is a new plug-in for After Effects and Digital Fusion:

http://www.algolith.com/products/algorithms/CAS/

Content Adaptive Scaling (CAS)

Description:

An autonomous and high quality Image Content Adaptive UP/Down Scaling (CAS) is proposed for any image scaling or pixel sampling rate conversion. Algolith's CAS is designed with emphasis for giving natural picture sharpness, minimal edge ringing and self-adjustment of filter coefficients for an arbitrary given Up and Down ratio.

Frederico: thanks for posting, but I chopped the main part of
since anyone can read it at the site you gave. It's also usually
not allowed to just copy whole pieces. A snippet to get people
interested is ok and if they want more information they can
follow the link.

I'm working on a DV <-> SD <-> HD up/down-sampler (with intelligent chroma upsampling for DV sources). What do you like / dislike about S-Spline? What features should be in my program. Bare in mind I probably won't have anything ready till the end of the year - I'm still in early development.

Graeme

the ability to adjust sharpness, and just geared towards video...avi output instead of having to externally convert to bmp and then back again

I am very excited to hear this! Keep up the good work!!
Rob

I'm a mac developer, so at the moment it's written in Cocoa, Quicktime, C, and OpenGL. If the results are good I'll get it ported to a PC platform.

At the moment, I'm working on still frame tiffs to test out the algorithms, but the finished product will be totally geared towards video. I already have a great de-interlacer and DV de-artifacter (intelligent chroma upsampler) as part of my filters for FCP, so I'll but improving on these for inclusion into the new stand-alone application. I'm also looking at sharpening (again, a new algorithm) and noise reduction.

I'll try and keep people informed of progress and show some test frames when I'm ready....

Graeme

As usual, you are working on something that is on everyone's minds.

My view on S-Spline is that it overly sharpens the image when it is resized. In my earlier post in this thread, I linked to two images of a cat that someone had done, and the resized image looked much sharper. Looks like the contrast is boosted and whites are blown out a bit. A caveat in the caption says that it was compressed for the web, but I wouldn't expect that to affect contrast in the way that it does.

I think a colorspace conversion is a good idea for this resize as well.

At the moment, I'm not adding sharpening - this method doesn't need it! But I'll want to add it as an option. Also, I'm not effecting the colour balance at all - it remains neutral, but again, I could put in full RGB curves if that's needed....

Graeme

I do a lot of digital still imaging via a Canon 10D. There's quite a number of sharpening algorithmns on the market, perhaps the best one is either Neatimage or FocalBlade. NeatImage is, principally, a de-noiser with sharpening technology added on. FocalBlade is a sharpener. Both pieces of software work quite well, however, they should not be applied until after uprezzing. Genuine Fractals is another very effective and interesting algorithms. Regardless of which one you pick, the per image processing time is quite long. I don't see the applicability to image sequences that are very long. At least, not with the current spate of processor technologies.

The sharpening techniques in those apps look like variants on unsharp-mask techniques which produce the traditional white and black lines around "objects" in the image, and although a little can look good, doing too much can look very bad.

The techniques I'm working on are for video, but I suppose they're equally applicable to photographic enlargements, and indeed, I'm working on still frames for my initial tests... Any sharpening in my processing does not use anything remotely like unsharp-mask and does not produce the black and white outlines. In fact, the processing I'm doing is nothing like any form of traditional image processing at all - you won't find it in any text book.

As for Genuine Fractals - I've seen examples of uprezzing using their software, and examples using S-Spline, and I'm totally confident that the results I'm producing look much better that these.

As for processing times - that can be an issue, but it's going to take a couple of seconds a frame to render, and that's with totally un-optimzed code. And if the quality is there - who cares about such render times, especially how processor power increases almost daily.

Graeme

The native LanczosResize function in AviSynth is pretty great for uprez - as it keeps sharp lines sharp... and best of all - it's free. Much better than any kind of Bicubic in the upwards direction.

Lanczos is just another interpolation filter. Neither BiCubic, Lanczos, Sinc etc. are intelligent in any way. The current push forwards in re-sizing technology is centred around an approach that is "data dependent" in that it relies in analysis of the image to produce a scaling that preserves smooth edges and lack of aliasing and jaggies that standard interpolation filters cannot. This is the area which I am currently researching, and indeed, having some good success. However, it's a long and complex route to take because, being data dependant, you must test your algorithm against many types of images to make sure that what works for one, will work for another. Indeed, my initial code worked great on organic images, but faired less well on cartoon outlines, whereas a standard interpolating approach will perform equally well with all types of images.

Graeme

Certainly... I only mention it, because the previous posts talked about the pros and cons of investing in various software... along with trying various things in photoshop (incremental bicubic uprezzing). I imagine that your software will be commercial based on your descriptions of its quality (sounds exciting!). The Lanczos in AviSynth is simply an alternative that is available right now, and can't be done natively in photoshop for people trying things without investing anything. It won't hold a candle to anything analytical or data-driven, but if all you've got is Photoshop - it's a different choice.

Absolutely - the intent of my post was to inform about the current research, not just the research that I'm doing - into image uprezzing. It's only going to become more and more important as people try to take their old SD formats and make them usable in a modern HD environment.

My research is going slowly, but it's an immensely complex, interesting and often infuriating field to work in! I'll certainly keep you all informed as to my progress.

Graeme

Lanczos algorithms for still image printing is also available from a printing product called Qimage. Qimage is especially helpful for uprezzing still images. Mike Chaney, the author of Qimage, has a few novel and interesting different uprezzing techniques he makes available. One of the more interesting approaches attempts to sharpen images that have lost rez due to the pixel color masks native to CCD sensor masks.

I read about Qimage as part of my research - it seems like a great program. Taking into account CCD masking isn't as important for people with 3CCD cameras, but it could make a big difference for 1CCD cameras. I don't know how my algorithm will cope with these differences yet as it's still in a really early stage.... If anyone has had some good success with other up-rez programs I'd be interested to know, for comparison purposes.

Graeme

I've got some more tests of my uprezzing algorithm to show:

http://www.nattress.com/magic.htm

Comments, feedback appreciated.

Graeme

Very exciting stuff. Looks gorgeous to me. Will this be available anytime soon in your commercial products?

Available yes, soon no... It's still very developmental, but it's finally reached the point where I can show it working on a real image. But next I have to get it working on video, optimised, tweaked, user controls, settings etc., so it's still quite a way off....

Graeme