Noise level survey

[High Speed Dubb]

I recently put an obscure but potentially useful feature into the Peach Smoother AVISynth filter which can give an objective readout on the noise in a video source.

The filter is here
http://students.washington.edu/ldubb/computer/PeachSmoother.zip
and the relevant option is “readout.”

Would anyone be up for running some card evaluations? I’d be very interested in seeing how the different input methods and tuners compare.

In case it isn’t clear, there are certainly some limitations to this. For one thing, noise isn’t the only criterion for picture quality. For another, the readouts are going to depend on card settings (especially contrast), interference, and source quality.

To get a good readout, use the command PeachSmoother(readout = true, dot = true)
(Dot isn’t necessary, but it will help confirm that you’ve got a stable estimate.) The estimation can work on normal broadcast video — You don’t need to use a test screen or a calibration disk. Just avoid video with constant motion, or clips which are especially dark.

[High Speed Dubb]

Just to start things rolling — My input card is a bt878a based Pinnacle Studio PCTV Pro. Using its Philips MK2 (NTSC) tuner for US Cable, capturing at 640x480i with luma peaking disabled, I get noise levels in the range of 5.4 to 5.8 (on stations without notable interference) up to 9.7 for one especially high interference station.

(The number is measured as expected absolute luma and chroma change per pixel in stationary areas.)

Using my Playstation over S-Video as a source, the noise level ranges from 0.95 to 1.2. (I don’t have a DVD player, so that’s the best clean source I have available.)

My card is calibrated to the full luma range [0,255], so my values should be elevated by about 256/220 over what you’d get when limiting values to the standard range.

[soulfx]

I've got an ATI 8500DV capturing at 720x480i over the my local US Cable TV feed. The only big problem I have is crosstalk. I usually get sources that come in with a noise of 4.1 - 3.5.

[^^-+I4004+-^^]

i have some VHS' so noisy your head would spin(fast)!hehe
i'll test it tomorrow
(also i have some clean and noisy analog satv channels...these tend
to have typicall sort of noise.....bigger black/white dots play around the screen....unlike noise from terrestrial broadcasts....
these dots are big and persistant and no temporal smoothing helped me
with those.....i should analyze if they(dots) stay on screen for more than a frame so tsmoothers get confused(?))

no tuner stuff from me.....i keep it strictly composite....
(and have no device with tuner)

stay "tuned" for more_

[High Speed Dubb]

The filter is designed to look for “white” noise. So it will probably ignore those persistent black/white dots, thinking that they come from motion.

[Richard Berg]

WinTV 401 fed by Time Warner Cable, coax snaking around several computers, top 2 lines of pure noise cropped out.

Good channel, still scenes: 5.0 / 3.8
Ditto deinterlaced with TomsMoComp(0,15,0): 4.0 / 2.9

All values stayed within +/- 0.2 at least 95% of the time. This was a video-based documentary piece chosen for its copious still scenes; though I've seen better image quality from (IVTC'd) film material, it's usually due to better lighting/levels control & detail -- noise should be the same, and easier to detect here.

[^^-+I4004+-^^]

Readout:[bt8x8 device at 704x576 & last two samples(9 & 10) at 720x576]
source--codec--readout--my(subjective)
estimation(1-5)

1_satv--huff--13/10--4(pretty noisy "5" would be noisiest)

2_satv--huff--4,1/2,7--1(i think i can't do less noise)

3_satv--huff--5,7/4,7--1,5(luma peaking on)

4_terr.tv--huff--7,7/5,9--3(noise+interference("grid")over the image))

5_terr.tv huff--6/4,6--3,5(noisy tv channel)

6_terr.tv huff--7,5/5,9--4(noisy tv,luma peaked)

7_VHS--huff--4,1/2,8--3(noisy VHS seems like Peach doesn't detect chroma noise...lots of it here)

8_VHS--huff--5,3/4,4--3,5(SAME VHS source luma peaked)

and now for some ultimate testing_
i will attempt to capture the same VHS recording
with both
huff & mjpeg....let's see how those two
compare noise wise
(more noise on huff is expected..,
but also more sharpness..)

9_VHS--huff--5,3/4--3


10_VHS--mjpeg--5,4/4,2--3

(looks like i was wrong-->perhaps more mjpeg noise can be
attributed to "quantisaton" (mosquito) noise which is off
course bigger in mjpeg than with
huff...mjpeg was at "20"...)

some additional thoughts-->
-on the satv analog noisy source you'll see (i'll provide screenshots from VD
"Copy source frame to clipboard") something that i called
"line tearing"....lines in analog signal are moved left or right from where they should be
i have come up with some explanation---->if noise "dots" (bigger b/w dots i mentioned eariler) "hit" the horizontal sync pulse of video then capturing device is confused and moves the line accordingly...that's the reason for this "line-offset" that occurs.....
noise can occur everywhere on transmission channel so it's natural that it occurs during the h.sync pulses too........

_i've noticed that peach doesn't measure chroma noise (like in VHS)
because i got pretty lo noise figure even though VHS source had much chroma noise.....but no,i'm not asking for this to be included in the filter itself
because chroma noise is easy to get rid of (CNR,etc.)

_once again i would like to thank Lindsey Dubb for providing such AWESOME
filter (it's not the documentation only..hehe)
i think it kicks c3d's ass seriously....
it's obviously done with more powerfull noise detection features,
"dot" and "readout" functions make it even more fun,it bares in mind that
too much smoothing leads nowhere....etc.
it must be that i'm in love!hehehe

_so as a final thought i'm wondering where did all the praise went?
c3d was considered a god when it came out,but now the REAL tool (for us analogue folks) is available everything is kept silent.....
(perhaps oversaturation with smoothers alltogether?)

and i guess i'll be using this fruit on my hi-res capturings....offcourse...

note-->i'll upload pix to my web-space,and as soon as i do i'll inform_

here it is_

http://free-st.hinet.hr/ikostic3/video_/video.htm

pictures complement the stuff written here,but pix can also be "used" for themselves (noise figures and green dot(s) are visible)
description "10" goes for image "Clipboard9-1" that's the only
difference...all other images' numbers correspond to thier descriptions....
there are 10 images.......

there...i hope this will be of some help for further explorations of analog
noise phenomenon.....and a ways to get rid of the noise while keeping image
as sharp
as possible (yes,peach is a way to go....)
[also i have much,much noisier VHS' but didn't tested it because i wouldn't be able to show pix as this is bootleg porn stuff....anyhow there's so much noise there that capturing itself is pretty tough too....on those tapes half of the signal is noise....hehe]

cheers

Ivo

[edited_formatting of the post(my ohh my does this such...can anyone
make some java code so that posts can be edited (on-line) as in real text-editors.....bah...)
added link to web space....note _there's interesting "side-effect" to temporal smothing (no this wasn't "peach") in conjuction with lobitrate one pass divx in "divx-samples_.zip"...in 3rd avi ,firecracker scene is really ugly...]

[High Speed Dubb]

So far it looks like soulfx’s ATI 8500 is the cleanest, though of course we’re not really looking at enough measurements to reach any conclusion.

Just for a clarification of what is being measured — The filter looks for the lowest level of noise on the screen, assuming that everything else is motion. So if the top couple lines are really noisy, they’ll just be ignored.

Richard,
Documentaries are often difficult material for noise measurements, since they’ll sometimes use low quality stock footage with completely different noise characteristics. But it sounds like the one you were watching didn’t tend to do that.

I hadn’t thought of running the measurements after another filter. Because the Peach tries to detect the lowest level of noise in the picture, it’s difficult to interpret the result. The reduction could be due to bobbing, since averaging pixels causes a reduction of variance. Or it could be due to the filter searching for pixels which are similar in both nearby fields, which also would reduce variance. Or it could be an indication that TomsMoComp is doing very well, and is accurately estimating the missing pixels. (Though that last one could well be true, I don’t think it’s the real reason for the measurement, since the motion compensation should only matter in moving areas.)

^^-+I4004+-^^,
The filter does consider chroma noise — Just not that kind of chroma noise. From the picture you posted, it looks like the noise is color crosstalk, since it’s only in the area of the high frequency letters. The area away from the letters doesn’t show those colors, so that’s where it is getting its noise estimate.

I’m glad you like the filter. But from what I’ve seen, the reaction to it has really been very good. It’s the Grape Smoother which seems to have been passed by.

[Richard Berg]

Documentaries are often difficult material for noise measurements, since they’ll sometimes use low quality stock footage with completely different noise characteristics. But it sounds like the one you were watching didn’t tend to do that.

True -- I meant something along the lines of 20/20, not Ken Burns.

I hadn’t thought of running the measurements after another filter.
With MoComp I agree it's more of an "interesting" result than an insightful one, but surely you'd consider something like Telecide(post=false).Decimate() helpful for Peach to determine the true noise of a telecined clip (rather than flagging on artificial combing)?

Would you like me to also test some "bad" channels?

[^^-+I4004+-^^]

HSD>So far it looks like soulfx’s ATI 8500 is the cleanest, though of course we’re not really looking at enough measurements to reach any conclusion.

_this is relative->the card that smoothens the image too much
will have lower noise figures too,
i think asus' owners would have tremendous results in "on-line" smoothing ie. denoising....with too soft picture offcourse....
if i go to bt8x8 tweaker and lower the luma peaking
(to "0" out of 8 states....."4" is default for bt8x8 chip)
then i might as well get better results than 8500.......
but then image is too soft and unacceptable.....
and my 4,1/2,7 is better than ATI's 4,1/3,5 isn't it?
butt off course this isn't any indicator for card quality etc.
every source has different noise.....
[but provide me with some ATI screenshots with some graphics (subtitles,station logo's,letters etc.)
on the screen and it's easy to say if the picture is sharp or soft]




but i guess that any deinterlacer will
lower the noise figure as it (to put it simple) smoothens the
picture to some extent.....


Grapes?hell we'll eat them too...(soon)...hehehe

[High Speed Dubb]

Richard,

If the material is interlaced, the filter still compares only between matching (top/top or notTop/notTop) fields. At least, that’s how it’s designed to work.

So for 2:2 pulldown material it shouldn’t matter very much whether IVTC is used first — either way you’re comparing corresponding fields. (The correlations would work a little differently if you join the fields first, but it isn’t clear which way would be better.)

For 3:2 material, things are more confusing, but it’s probably still best to run the Peach first, at least for noise measurement. That way you get the benefit of one pair of fields between which there is definitely no movement, providing a very nice measure of the noise. On the down side, it does mess somewhat with the motion estimation.

PS: I don’t think the bad channels are going to be as much use to other people. Their values could have more to do with interference than the card. Then again, maybe some cards are less prone to interference.

^^-+I4004+-^^,

Yep, I didn’t mean to imply that the ATI is the best card, just that it has the least noise of the bunch which have come up. Like you say, a blurry picture will tend to have less noise. By this measure, a standard VCR would probably beat S-VHS.

4.1/3.5 is probably (barely) better than 4.1/2.7, since it implies a more uniform amount of noise in the picture. (That’s why I just posted the NoiseLevels — The baseline can be a little confusing.)

[High Speed Dubb]

Using my same (bt878a) card, here’s what I get for my Philips VR1010 VCR...

On S-Video input, I pretty consistently get noise which is just slightly (0.1) higher than from the card’s tuner.

On composite from the VCR, the values seem to be consistenly about 0.1 higher than S-Video (so 0.2 higher than the card’s tuner).

The VCRs picture appears to be sharper — overly sharp, in my opinion.

[soulfx]

I turned off the Noise Filters with the 8500DV. For default it has them cranked all the way up for Television capture.

[soulfx]

Here are some screenshots. As mentioned above, I have Noise Reduction Disabled as well as Adaptive De-Interlace and VPE Adaptive Disabled. With all these settings on (default) the picture is a bit on the soft side.

I've made sure to get a good S/N ratio input into my Tuner as these 8500DV's are really picky on their input Signal. I've just got one spliter on my the cable for my modem and then rest of the ends are capped. Whenever I get around to it I'll run cable to the rest of the house, but I want to make sure I don't lose Signal.

One thing that I can't control, but only complain about is the quality of the local affiliates around here. There is one station that really lays heavy on temporal smoothing (you can see ghosting artifacts even on a regular TV it's so bad). The same station also has problems with color and station identifier overlays (AGH! I hate that station, I really have to call them up and yell at someone).

Anyways here are the screenshots. It's just two, but you can get the idea.
http://members.cox.net/soul.fxpro/8500DVscreenshots.rar

I was using Peach after Telecide. When I put it before, I just got a little bit lower noise estimate, but not by too much only within .1

[^^-+I4004+-^^]

soulfx,this looks rather nasty!
looks like you're using overcompressed mjpeg(?)
this is not good source to begin with.......

why haven't you included noisefigures in the pix?
(from the VD input(asynth feeded) window or something?)

have you seen my samples?
yes they were huff,but even for mjpeg i can get much sharper image etc.
i can't believe that you US folks get this lousy picture quality...
quickly,someone post some nice 640x480 samples (hehehe)

and the graphics (qvc) are soooo smeared.....

at first i thought you captured at 320x240 and upsized..(heheh)

and now something more constructive

wouldn't it be beter to reduce color saturation a bit,and to
decrease mjpeg compression to "18"-"20"?


and deconstruction again hehe(don't tell me this is huff as then i'll be very happy i don't live in US...hehehe)

anyway thanks for the insight from US analog capturer perspective.....

[soulfx]

I just clicked the screenshot grabber in MMC so the images you are seeing are captured purely from the Tuner Source.

[^^-+I4004+-^^]

then perhaps i should of used DScaler instead of
real captured avi?
(heheheh)
why is there so much mosquito noise?
is that .png's stuff,or?
wouldn't it be better to take hi-quality bmp
and jpeg optimize it?

yeah...well.....
as they say gift is a gift...

[High Speed Dubb]

You think those QVC graphics are smeared? I forgot to include QVC in the noise measures from my system, since I had removed it from my channel list. But if I turn it on, the noise level there is an amazing 20.1. (By the way — to get accurate readouts at that high a noise level, you need to increase the NoiseReduction setting. It shouldn’t matter for video with more normal noise levels, though.)

PNGs are lossless, so we’re seeing the real image. It does look a bit blurry to me, too. The card may be stopping the luma at a bit too low a frequency.

[^^-+I4004+-^^]

my QVC (or rather HOT,but it's the same thing) vidcaps are in thread "Tested capture cards" that i
started
here it is:
http://forum.doom9.org/showthread.php?s=&threadid=32516

take a look at stuff i attached on my second post to that thread,and compare that to soulfx's stuff.........[636kB]

PNG's are lossless,but then where does quantization(mosquito)
noise come from?
or is it that ATI suddenly got even worse?
i guess mjpeg was selected as an input format?
i don't think it's right to do screenshots this way..it's sloppy work if you ask me....
we should try to get as close to the image quality that is in recorded source .avi etc.!

soulfx?


i'm not sure,but i think that with elsa (erazor3) at 352x288 and then upsized i can get better pictures than those....
(i would have to use overlayed (antialiased) image as a source for upsizing,so i wouldn't use huff,but elsa own capture codec etc.)
if there's any interest i'll do this test too.....
(yes it's hillariously poor quality on those samples of soulfx....)


also Lindsey,you say that that you turn on NR with hi levels of noise?
what's the use then?
isn't it source noise we're measuring?
but ok,with such high noise levels i doubt any capturing (and any subsequent filtering) will
look good etc.

i can't believe that NTSC lowered resolution (cca 480 vs. cca 570 of PAL) results in such lowered quality!
but i guess with proper (huff) capturing and some bmp->jpeg conversion (for posting purposes) NTSC capturing can look much better....
Lindsey do you agree?

[soulfx]

I thought that screen captured directly from ATI MMC would look the best, but I guess I was wrong. I wanted to get one before Peach took out the noise or any type of compression was applied to it. Everyone is right about the fact that it does look like it applies somekind of compression because it just doesn't look right at all. One thing about this card is the lack of really good technical information and support.

I've pulled some frame grabs from my recent Bernie Mac record and put them up in place of the old ones. They are pulled directly from VDUB and has Peach enabled with Estimates on.

I'm sort of not worried too much about the quality of this card right now. I'll just have to hold out for another month or two and then I'll get a HDTV card. The local stations around here have been working hard at getting their transmitters upgraded for OTA HD broadcast. I'm just waiting for two stations that said they would be up by the end of this month early next month.

Here are the new framegrabs:
http://members.cox.net/soul.fxpro/8500DVscreenshots.rar

[High Speed Dubb]

The original pictures didn’t look compressed to me. The hanging dots on the bottom/top of the green areas are artifacts of a 2D comb filter. (You can also see that in the Bernie Mac logo.) The stuff around the text is crosstalk.

Since we’re trying picture quality comparisons, here are a couple fairly detailed shots from my Bt878 card. They’re taken from DScaler, using the equivalent of Peach Smoother, Guava Comb, and GreedyHighMotion. From what I’ve seen, DScaler and capture results are the same.

(These pictures are each about 500 KB.)

http://students.washington.edu/ldubb/computer/images/News.png
http://students.washington.edu/ldubb/computer/images/Hardware.png
http://students.washington.edu/ldubb/computer/images/Nuts.png

The inaccuracy in the NoiseLevel estimate with low NoiseReduction and high NoiseLevels doesn’t hinder the filtering — in fact it makes it more accurate. The only problem is that you can’t interpret the Readout result as an expected absolute change per pixel. Instead it gets you (absolute change per pixel)*K, where K grows slowly as the real absolute change increases over about 12. (That’s for a NoiseReduction of 35.)

About NTSC and PAL — They’ve both got their points. NTSC has the refresh rate, PAL has the resolution and better (though still poor) color accuracy. PAL comb filtering is more difficult, but PAL does a better job of hiding uncorrected crosstalk. For my own purposes I’ve decided to view NTSC rather than PAL — because that’s what the stations around here broadcast. :)

[^^-+I4004+-^^]

[as i dload your samples with dial-up]

LD>PAL comb filtering is more difficult, but PAL does a better job of hiding uncorrected crosstalk.


_what do you mean by "hiding uncorrected crosstalk"
as far as i know PAL doesn't have a system to hide
any crosstalk between luma/chroma......
you should see how color kicks in on the tiny,fine
patterns,but obviuosly both systems have it (except for secam
maybe...secam
modulates color differently to PAL and NTSC....)
it corrects phase changes (you have hue control,we don't)
but separating luma/chroma completely?
i think this can't be done properly as luma and chroma overlap...

also,i see work on DS has begun again.....
timeshifting?
but i guess on line deint.+filtering +capturing
can't be expected soon?
(i tried it on lo-res (DS 3.something) and i must say i was surprised how good)
the picture looked.....)

pictures?
aha,now it's MUCH better!
[that "head"(on "Hardware.png") looks bigger than we have here....and there's two of them on V machines.....LOL i'm saying this and you probably don't know a thing about it as you signed pic "hardware".....hehehe]

cheers

Ivo

[High Speed Dubb]

as far as i know PAL doesn't have a system to hide
any crosstalk between luma/chroma......

It does. You get opposite phases in adjacent scanlines in an interlaced frame. Since the opposing colors are next to each other, they’re supposed to blur together a bit, reducing the effect. With NTSC the corresponding scanlines have the same phase, which makes the dots in dot crawl easier to see.

Theoretically you can’t get rid of all crosstalk, but you can get pretty close in most circumstances. The advantage of NTSC for this is that you can use adjacent scanlines to cancel crosstalk, while for PAL you need to skip one scanline, making it more susceptible to detail in the picture. There are similar factors when using a temporal comb filter.

The picture named hardware really does have hardware on it. (It’s from a car show — remember that hardware doesn’t just refer to computer parts.) The nuts one, on the other hand... ;)

[^^-+I4004+-^^]

LD>It does. You get opposite phases in adjacent scanlines in an interlaced frame. Since the opposing colors are next to each other, they’re supposed to blur together a bit, reducing the effect. With NTSC the corresponding scanlines have the same phase, which makes the dots in dot crawl easier to see.

_proper PAL decoding--->scanlines 1 & 3 give "new" line 2
3 & 5 give new line 4
5 & 7 give new line 6
etc.

for example,scanline 1 is slowed down 64us so it can be added to
line 3.......this is proper PAL decoder and is called "pal-d"
(pal demodulator,or pal+delay line)
adding these two lines you can cancel the phase error
(2 lines each with same phase error added give one correct-phase line)

compared on b/w receiver,you could notice one line downwards(to bottom of the screen) shift on color version of same signal
(if you look same PAL signal in paralel on 1 color and 1 b/w tv)

but yes,this PAL stuff is not simple to grasp,it's by far the most
complicated color transmission used and one can be happy if it works
at all....and it does and nice........



stuff you described is so called "simple-PAL" and it doesn't have delay line -->it projects both lines on the screen and counts on the effect of merging colors in human eye.....
this is bad way to do it and i have never seen tv-receiver with this
(but i understand they were in existance on the beginnings of the PAL as not every factory could made proper PAL demodulator system....)


this has no effect (in my mind) to "hiding crosstalk".....
(opposing colors are NOT next to each other in proper PAL system!
now one chip and one delay line are all is needed to make proper pal decoder! and virtually every PAL receiver has proper PAL decoder)
crosstalk must be adressed in ONE line (as it happens in ONE line)
it has nothing to do with "merging lines" etc.

if there's hi-frequency luma (above cca.3,5mhz or so) it will disrupt the demodulator..it'll fool him to believe it's chroma (as these are separated by mere notch filter) and it will show him on screen as chroma!(which off course is wrong!)
this is so called "cross-colour"
simillar occurs when chroma gets to be decoded as luma and disrupts the system but these effects are much smaller and less visible (so called "cross-luminance" effect........coloured edges are affected...)

so anyway cross-talk should be divided into these 2 categories,as
they are not the same
(cross-colour and cross-luma.....cross colour is more visible)

NTSC?well i don't receive any here (heheh)
i know that phase changes mean colour changes too(as color information is carried by amplitude AND PHASE of color subcarrier
in PAL it's 4,43 mhzs in NTSC 3,58(?).........)


LD>The picture named hardware really does have hardware on it. (It’s from a car show — remember that hardware doesn’t just refer to computer parts.)

_i was reffering to the name of that big part on the desk....
it's a head of an auto engine....as in "2 overHEAD camshafts" etc.


cheers_

[High Speed Dubb]

^^-+I4004+-^^,
That was awfully long for an off-topic post. Please try to stick with card noise assessments for this thread.

[seldon]

Hi there

My setup:

Capturing with my Terratec Cinergy 400 @ 640x480 in iuVCR using huff. All clips were interlaced.

Playing the following script in vdub:

LoadPlugin("D:\capture\avisynth\PeachSmoother.dll")
AVISource("D:\capture\iucapture.avi")
SetMemoryMax(40)
PeachSmoother(readout = true, dot = true)


I tried capturing from 10-15 channels and the results are as follows:

All the captures start out with a value of 4.1-4.2

2 of the channels stay very close to this for the full clip.

The rest of the channels drop rapidly to a value between 2.9 and 3.3 after a couple of seconds. The value stays in this range for the rest of the clip, occasionally peaking at 3.5 on some of the channels.

It seems to me that the noise readout rises with increasing motion.

It also seems to me that the non domestic channels give a higher noise level. Which is very likely considering the crappy cable provider i have.

-Seldon

[fisix]

hmm,

Lindsey, you could also put in code that would register other picture quality factors, and then the noise characteristics might be more comparable from card to card.

otherwise, using a composite or svideo input from a standalone dvd player as source might be a reasonable standard.

my wintv pvr-pci bt878a based card doesn't seem to have noise issues so much as an overall gain and clip issue that bleaches out most light colors and offsets the others. i'm not setup at the moment to test clips or i'd have a number for you. maybe we should all test caps of the simpsons?

[High Speed Dubb]

seldon,

4.1ish is the starting default noise estimate — It doesn’t tell you anything about your reception. The filter places very little reliability in its initial value, so it should quickly move from there to an actual estimate.

So your 3ish values are the real estimates. It sounds like you’ve got a nice, clean signal, and/or a good receiver card. How do you like the picture quality? It would be cool if we could spot some standout tuners.

The noise evaluation works by using some heuristics to tell it whether the estimate is accurate. At the start of the run, just about anything will affect the noise estimate, since it hasn’t yet seen a decent signal. After a still scene, the filter is able to have much more confidence in its measurements. But even after that happens, the noise estimate can rise very slowly when there is slow motion or handheld camera shots. (Fast motion is easier to reject.) I’ve got some internal parameter tweaking in the works to make it faster to adjust and slower to change thereafter.

fisix,

It would be great to have a way to measure other aspects of picture quality, but I don’t know how to do it. (Well, I can think of a way to evaluate comb filters, by comparing composite vs. S-Video from the same source.) I’d be up for any suggestions, though.

Yeah, results from an external DVD (or console, which is my best clean source) would be informative, since that would let us separate signal/tuner issues from other sources of noise.

I’ve had trouble with the gain on the Bt878a, too. Calibrating each station separately helps a lot, but brighter colors still aren’t differentiated as well as medium or dark colors. The newer DScaler builds have a histogram filter which makes manual channel calibration somewhat easier.

I like the idea of having a program all of us use to compare. It wouldn’t be perfect, since you’d partially just get an evaluation of your FOX carrier. But it would at least mean the noise in the original material would be the same.

[Roginator]

I just got a capture card based on the new CX881 chip on Friday. It's called the XCapture and sells for $39 (plus $13 shipping).

I tried the test mentioned in the original post to this topic and got the following: 0.963 and 0.681

Nice, eh?

This was on a 320x240 clip recorded from Showtime from a digital cable box and through a cheap VCR since there's no tuner on the card. Composite was used since the cheap VCR doesn't even have s-video. No compression on a 2 minute sample.

There is one problem -- I don't yet have a program which will capture in avi which allows multi-segment captures. I ordered iuVCR in hopes that will work. I used VirtualVCR to do the test. (Note: iuVCR works after adding the /notv switch to the shortcut)

The program supplied with the card allows lengthy MPEG captures, but I'd rather get full quality.

[seldon]

High Speed Dubb,

4.1ish is the starting default noise estimate — It doesn’t tell you anything about your reception. The filter places very little reliability in its initial value, so it should quickly move from there to an actual estimate.

Ok.. so I guess it's just a freak accident that two of the channels stay precisely in that range.

So your 3ish values are the real estimates. It sounds like you’ve got a nice, clean signal, and/or a good receiver card. How do you like the picture quality? It would be cool if we could spot some standout tuners.

I have no idea how clean my signal is compared to other cable providers or other contries. I have very limmited knowledge about the quality of the reciver card compared to BT based cards, but I think Swan once stated that her impression of the picture quality compared to a BT based card was in favor of the philips based(Am I right Swan?). This might be due to the 9bit A/D converter on the philips card, but someone please correct me if I'm wrong.

Anyway.. I'm very satisfied with the picture quality from the tuner. I've never tried using the composit or S-VHS connections to capture.

-Seldon

[Swan]

Originally posted by seldon
but I think Swan once stated that her impression of the picture quality compared to a BT based card was in favor of the philips based(Am I right Swan?). This might be due to the 9bit A/D converter on the philips card, but someone please correct me if I'm wrong.
You are right, Seldon. But I was possibly saying this because I compared a card with the Philips chip with a *faulty* card with Bt878. :)
I just discussing this in a different thread an I'm lazy enough to virtually just paste it here instead of re-phrasing it. :)
-----------------------------------
Initially I thought it was incredible; the difference between the AverTV Studio card I had (bt878) and the Cinergy (SAA7134) was that big! Of course, I could only *watch* TV with it, not capture, so I was never able to form an opinion on what full res. captures made with the Cinergy looked like. After I bought my present card, the Pinnacle PCTV Pro(bt878), I noted that the Aver card was faulty, that's why I thought the Cinergy had such incredible quality. The Pinnacle produces excellent results. I can't say I saw any major difference between the quality of the Pinnacle and the Cinergy (when watching TV).

I have given it to a friend and when he begins to find his feet I'll be able to see some samples of 720 x 576 captures from the Cinergy. Then I will be able to say if I think the Cinergy's quality is better or not, compared to the Pinnacle.

Thread here: http://forum.doom9.org/showthread.php?s=&threadid=37701
/Swan