Noise comparison: 35/4:2:0 vs. 180/4:2:2

[Dan Keaton]

Dear Luben,

I just posted an answer for you in the other thread.

http://www.dvinfo.net/forum/converge...1-5-249-a.html

[Piotr Wozniacki]

Just to get back on topic here...

Previously, I have posted a comparison of EX1 native vs. nanoFlash I-Fo/220. It showed the compromise one is making when using the 220 Mbps, I-frame only mode - *some* increase of noise, but at the same time *considerable* decrease of the mosquito noise when compared to the corresponding EX1 native picture...

Well, to show a similar comparison on the other extremum - the nanoFlash Long-GoP, 50 Mbps/4:2:2 vs. the native EX1, 35/4:2:0 images - I'm posting the 2 grabs below. One can clearly see that while the noise is *not* increased, the detail definition (see the roof, and the foliage) is much better in the nanoFlash picture - in spite of its bitrate being just 50 Mbps!

Heck - the improvement in the lower-left roof detail is even visible in the attached thumbnail view!

[Dan Keaton]

Dear Piotr,

The nanoFlash image is labeled 35 Mbps. Is this correct, or is this a 50 Mbps image as you described in your post?

[Piotr Wozniacki]

Yes indeed - my mistake, sorry. It IS 50 Mbps - I'll try to correct this...

PS. The label's now correct.

[Piotr Wozniacki]

I couldn't resist - here is another comparison, again between the EX1 native and the nanoFlash L-GoP, 180 Mbps. This time however, these are full frames - not blown-up crops. Just notice the much better edge definition of the red flower that the 180 Mbps nano clip is giving thanks to the 4:2:2 color resolution...

Scenes like this do not contain large, uniform color, shaded areas that could introduce noise - this is where the nanoFlash really shines!

I guess this posts nicely rounds up this thread, as it goes back to the "35/4:2:0 vs. 180/4:2:2" subject...

[Piotr Wozniacki]

I've been testing my EX1/nanoFlash even more, trying to find a combination of settings to both that would minimize the noise - and thinking...

So far, the only post (in this thread or in the one at EX forum: http://www.dvinfo.net/forum/sony-xdc...ctive-ex1.html) that agrees with my observation fully, and doesn't indicate I'm splitting hairs, or using improper comparison methods (like posting jpeg's instead of png's), has been the one by Voitto (http://www.dvinfo.net/forum/sony-xdc...ml#post1539612).

I've been wondering:

Is the reason for people not confirming the problem, that they don't want to admit it exists? You know, having spent a lot of cash on an EX1/3 camera, in some cases also on the nanoFlash - it can be sort of difficult to admit openly it doesn't perform as hoped, and expected basing on the nice 54dB S/N figure in the specs...

I could understand such reluctance, but I'd once again ask those knowledgeable EX users to step out and say openly what they think on the subject that's bugging me, like:

- either: yes, the noise is there, and in some cases - especially with high bitrates nanoFlash recordings - it can make the footage unusable without de-noising in post (which in my opinion, can negate the very sense of using high nanoFlash bitrates in the first place)

- or: no, I do not find my EX camera recording particularly noisy (and the higher nanoFlash bitrate I'm using, the better results I'm getting)...

Some feedback like this would help me tremendously - at the current state of affairs, I almost start to feel paranoid... Is my camera noisier than others, and should it be sent in for servicing? Or maybe my SDI cable is to blame?

'Cause you see, the only other explanation of seeing things other people do not see is that I'm editing my stuff using a large, 50" plasma display, hung above my desktop monitor, and watched from 1-1.5m. This way, I'm really seeing individual pixels, and can tell noisy picture from a clean one.

But, isn't using large displays what HD is about?

[Gints Klimanis]

Piotr, I am very happy with my Nano purchase. I moved from Sony Z1U to EX1 and was pleased but not satisfied until the Nano. I see a large quality increase in real detail and a substantial reduction if not elimination of macroblocking artifacts for handheld operation. The noise increase you describe is seen in my footage, but I am not as worried about noise as you are. I would prefer to record my scene and camera noise as accurately as possible so that future noise reduction technology operates on the mildly-compressed Nano files.

Have you tried the economical NeatVideo noise reduction plug-in for Sony Vegas? I will try that.
Neat Video :: filtration examples

[Piotr Wozniacki]

Hi Gints; thanks for the response.

I have just used the (free) Dynamic Noise Reduction filter from Mike Crash on a particularly noisy, 100 Mbps L-GoP nano file, and the results are promising.

People at the SCS forum say the Neat Video plugin is even better - I will definitely be purchasing it soon...

[Dan Keaton]

Dear Piotr,

Here are a few thoughts and questions that I have:

1. Are you doing everything possible to minimize the noise at the camera?

For example, are you using zero (or lower gain)?

Are the scenes properly lit?

Many, but not all of your examples are zoomed in, portions of images, in the deep shadows.

Are the controls in the camera set to minimize noise or to accentuate it?

2. When you record, using the nanoFlash at very high bit-rates, I strongly feel that we can faithfully capture the images that are provided to us.

If you can test your camera's output with a professional monitor you will be able to see if the camera has excessive noise or not.

3. Consumer televisions, even large screen plasma televisions, can have noise.

Personally, I have been testing some 23" to 25" LCD televisions, major brands, I was amazed at the high level of noise. When I play the same footage on my professional 24" Sony LMD-2450wHD monitor, there is no visible noise.

4. Viewing a 50" Plasma at 1 to 1.5 m is certainly not the recommended, nor typical, viewing distance.

How is your signal getting to your Plasma, is it via HDMI, Component, or via a computer signal via a DB-15 or other connector?

5. Do you see noise, when using zero gain, in a well lit scene, playing at normal video speeds?

6. As far as the 54 dB noise figure, I am assuming you are referring to the camera specs.
This value was probably achieved using far more light than the shadow details that you have been posting.

7. We have tested the nanoFlash, end to end, with a very nice Video Clarity system. The results are posted on our website, near the bottom of this page. These tests eliminate the camera variables and test the quality of the recording/playback.

Quality | nanoFlash | Video Recorders and Converters

[Piotr Wozniacki]

Dear Dan,

Thank you for the response to my doubts - not the first one in this thread :)

I'd like to assure you that - with my experiments (however unscientific they may be) - I have practically eliminated the "my nanoFlash being defective" unknown from the equation. The EX cameras definitely DO generate excessive noise (some people confirmed it for me in http://www.dvinfo.net/forum/sony-xdc...ctive-ex1.html "parallel" thread of mine at the EX forum). And to answer your question: yes, it can be noisy even in well exposed scenes, and gain at -3 db - this is what worries me most.

The nanoFlash merely makes it more visible (along with better details), when you compare its high bitrate picture with the native, 35 Mbps/4:2:0, EX1 compression.

What I'm trying now to achieve through my experiments is find the best combination of the on-camera settings (influencing noise), and the optimum nano format (particularly bitrate), to use in specific situations.

I'm also considering using de-noising filters in post, but somehow I still tend to think of it as the last resort... Not sure yet whether or not I will include de-noising in my post arsenal on a permanent basis.

As to the monitoring equipment I'm using, and the way I use it - well, I agree a consumer plasma is only what it is - far from the professional stuff. But I can assure you that when I manage to produce a near-perfect piece of image, it can look properly (i.e. perfect) even on the device I'm using - and from such a close distance, too.

And last but not least, a word on the examples I posted here: as you have noticed, I always illustrated my point with a pair of screen grabs from my NLE - one always being the native EX1, the other from my nanoFlash. Thus - even wiith a somewhat extreme examples, like blown-up crops, or intentionally dark scenery - it was the difference between native EX, and the nanoFlash, that was of interest; this way I guess there was nothing wrong with this methodology, even though it was far from scientific.

However, I have come to the conclusion that illustrating this kind of issues with stills is basically inappropriate in that in the moving video, the noise artifacts are shimmering, thus becoming even more noticeable - and NO still screen grab can show that.

While I'm typing this on my PC, Vegas is rendering a particularly noisy clip from my nanoFlash, on which I'm testing various settings of a NR filter. My purpose is to check whether the noise can be minimized while keeping most of the picture improvements that recording with the nanoFlash at 100 Mbps brought about...

I wish I had an upload Internet connection fast enough to post this clip somewhere for you to see - this is a typical example of a well lit scene, shot with noise-reducing settings of the EX1 (like the gain at -3 dB, crispening in the positive territory, etc.), and yet it contains areas literally swarming with noise!

[Tim Polster]

Piotr,

I have been following your threads and seeing how they turned out. This is how progress is made and how you get to know your equipment. This is a good thing and thanks for your efforts!

Now, I would say with my experinces with HD cameras, they are all going to seem a bit noisy under $20,000. I think it is a combination a chip technology and increased resolution for noise to appear.

Remember, we are still early on in the HD camera development cycle. The EX-1 was a ground breaking camera. So we have some improvement to be had as time goes forward.

Noise seems to be a much larger issue with my panasonic cameras than the EX-1. But the Nano has in my view really helped the noise situation on the HPX-500 as the Nano captures more detail allowing me to dial down the in-camera detail settings. This results in a cleaner signal going to the Nano which I then put higher quality sharpening on in post.

But on all of the cameras, I notice that the chips are a hair away from falling apart if light levels fall off. Basically, the HD chips and processing needs a lot of light. I have also noticed that the HD cameras have turned the "protect the highlights" style of exposing/shooting upside down. I find that I will get a better result with recovering overexposed images than underexposed images in post. Why? I think it relates to this noise issue.

To my eyes, the EX-1 is a very clean camera for $6,000. But I most often shoot in 720p60 so there is less real estate to show noise. On the Panny's the detail setting and the coring setting are the two noise controllers. I don't know what has the most influence on the EX-1. I guess that tells the whole story as I have never seen the need to eliminate the noise from my view on the EX-1.

So I would say keep on experimenting. My friend calls it "the sickness". and I have it too. But at the end of the day, you will get the best image you can get from your equipment.

Can you post some raw footage on a website that I could download and run through my system? I would be happy to compare to my footage.

[Piotr Wozniacki]

Hi Tim,

Thanks for the encouragement. I guess "the sickness" is just a less gentle term for "perfectionism", but it describes the very same phenomenon in my case :)

Thanks for you offer of inspecting some of my raw footage; what length clips (in seconds) do you think would be enough for me to upload (as I mentioned, my Internet connection - particularly upload - is low speed)?

Cheers

Piotr

[Piotr Wozniacki]

Originally Posted by Tim Polster

I have also noticed that the HD cameras have turned the "protect the highlights" style of exposing/shooting upside down. I find that I will get a better result with recovering overexposed images than underexposed images in post. Why? I think it relates to this noise issue.

Coincidentally, the clip I mentioned earlier today (the one I had to use the NR filter on), has been shot with exactly the above philosophy in mind, i.e. on the verge of blowing out the highlights. It was my intention to let as much light in the camera as only possible without unrecoverably loosing highlight details, and then return to the correct levels in post. I even engaged Cine4 to stretch the lower midtones even more on the foreground subject. BTW, it turns out I've overdone a bit with the exposure - there are fragments of sky that are burnt out, and detail cannot be recovered by mapping the picture back to the proper level using color curves. But this is irrelevant to the main issue - the noise:

Because-surprise, surprise: the noise I got in some usual picture areas (mid-IRE, uniform color) is even more pronounced now, than in the same scene exposed traditionally - i.e. with just traces of zebra at 100% here and there. Go figure...

[Piotr Wozniacki]

Originally Posted by Piotr Wozniacki

I'd like to assure you that - with my experiments (however unscientific they may be) - I have practically eliminated the "my nanoFlash being defective" unknown from the equation.

Dear Dan,

Unfortunately, I must retract the above statement. The nanoFlash encoder is still suspect - perhaps even the most suspect one in the hardware chain I'm using. Let me explain.

It's obvious and expected that any Long-GoP encoder will produce frames of varying quality within each GoP. For instance, the position and color of individual grains in the noise pattern change from one frame to another. The degree to which they differ determines intensity of the "shimmering" effect as perceived when watching video in motion.

Now, the problem with my nanoFlash Long-GoP encoder is that the higher the bitrate, the bigger those fluctuations between individual frames within any given GoP.

To exclude any other variables, the clip I've been analysing today contained exactly the same scene I posted previous examples of (the wooden shed), but this time I brought it out of the shade by engaging Cine 4 gamma which stretches blacks, and exposing so that the shed is much brighter (at the expense of the sky behind it starting to blow out, which is a limit for any practical shooting). Also, I shot the scene from the tripod, so the image is 100% static.

Then, I enlarged a cropped fragment (620 pixels wide), and studied *both* the native EX picture, and the 100 Mbps nano picture, frame by frame.

While in the case of the native EX1 encoding I could see *some* fluctuation to the noise pattern, they were very slight indeed; however - watching the nanoFlash 100 Mbps L-GoP picture - I noticed that some frames within each GoP are extremely noisy (see the grab to the right, below). In fact, those most noisy frames show some pixelation, similar to the example I attached to the post #103 we discussed earlier. The frames immediately next to those really bad ones, can be much smoother - as the attached grab to the left clearly show.

It is this strong fluctuation in the noise pattern and intensity which makes the actual video from nanoFlash shimmering - to the extent of being almost unusable (unless I run a NR filter on it in post).

Dan, I realize this thread has become very long and complicated one by now, but I hope you understand it took me some time and effort to reveal the true nature of the problem. Out of the hardware elements used, which is most probably responsible for the excessive shimmering noise my nanoFlash produces in L-GoP, high bitrate video? Here is my reasoning:

- it cannot be the camera sensors, as the native EX1 doesn't show such a strong fluctuation of noise between neighbouring frames within GoP's

- it cannot be the SDI connection, as there is no reason for it to fluctuate at all

- having eliminated the 2 above ones, it's most probably the nanoFlash encoder which is responsible. The only question remaining is whether it's just my unit, or all of them...

Dan, I'd like to put the clips I'm comparing here on a DVD and ship them to you for analysis. Is it OK?

Piotr

NOTE: the pictures below are NOT EX1 vs. nanoFlash comparison this time; instead they show 2 directly neighbouring frames in the same, 100 Mbps, L-GoP video from my nanoFlash!

[Dan Keaton]

Dear Piotr,

The Sony Long-GOP consists of I Frames, B-Frames, and P-Frames.

They are not all created equal in terms of quality.

If you want the highest quality, we recommend 220 Mbps or 280 Mbps I-Frames Only.

Our 280 Mbps I-Frame Only has been thoroughly tested by a major network for one of their high production value prime-time shows using the most sophisticated of test equipment.

The quality met their specifications and needs.

As I calculate it, you are looking at two frames, zoomed in 310%.

Please feel free to ship us the DVD, if you wish.