Sign in with
Sign up | Sign in

Test Class 2: Noise And Artifact Reduction

Video Quality Tested: GeForce Vs. Radeon In HQV 2.0
By

This class is all about improving poor-quality, noisy video. Today’s video processors can remove unwanted noise from video without noticeably blurring or degrading the picture.

Chapter 1: Random Noise Tests

Random noise is the grainy, changing pattern you might notice on poor-quality video. This chapter contains four tests, comprising four different video clips that present unique challenges for random noise removal. A perfect score of five is given for each scene if noise is reduced, but details are intact. The score is lowered to two if noise isn’t significantly reduced, and a score of zero is given if noise isn’t reduced or the details have been softened and smeared.

This test is fairly simple to assess, even if there is an element of subjectivity here. All of the graphics cards we’re testing achieve a perfect score, although we note that the Radeon noise reduction appears superior. GeForce noise reduction isn’t quite as strong when it comes to moving objects, no matter what level the de-noise option is set to.

Random Noise Test Results (out of 5)

Radeon HD 6850Radeon HD 5750Radeon HD 5670Radeon HD 5550Radeon HD 5450
Sailboat5
5
5
5
5
Flower
5
5
5
5
5
Sunrise
5
5
5
5
5
Harbor Night
5
5
5
5
5

GeForce GTX 470GeForce GTX 460GeForce 9800 GTGeForce GT 240GeForce GT 430GeForce 210
Sailboat55
5
5
5
5
Flower5
5
5
5
5
5
Sunrise5
5
5
5
5
5
Harbor Night5
5
5
5
5
5


Chapter 2: Compression Artifacts Tests

Compression artifacts are very different from the grainy patterns in random noise. Compression noise usually manifests as blocky edges on objects or even squared-off areas of the video that contain nothing. This chapter has four test videos that suffer from compression artifacts: a moving text pattern, a roller coaster scene, a Ferris wheel scene, and a large bridge scene with a lot of slowly moving traffic.

To achieve a full score of five in each scene, compression artifacts must be significantly reduced, while the details remain crisp. A reduced score of three is given if the artifacts are substantially but “not fully” reduced, and zero is earned if the artifacts are not lessened or the scene is smeared and detail is softened.

We find the compression artifacts test (and the upscaled compression artifacts test coming up) to be the most difficult tests to judge. The first problem is that the guidelines are quite subjective: five points for “significantly reduced” versus three points for “reduced, but not fully.” The other issue is that while certain cards are able to reduce some compression artifacts, the result isn’t dramatic. The changes are subtle at best.

Compression Artifacts Test Results (out of 5)

Radeon HD 6850Radeon HD 5750Radeon HD 5670Radeon HD 5550Radeon HD 5450
Scrolling Text
5
5
3
3
3
Roller Coaster
5
5
3
3
3
Ferris Wheel
5
5
3
3
3
Bridge Traffic
5
5
3
3
3

GeForce GTX 470GeForce GTX 460GeForce 9800 GTGeForce GT 240GeForce GT 430GeForce 210
Scrolling Text0
0
0
0
0
0
Roller Coaster0
0
0
0
0
0
Ferris Wheel0
0
0
0
0
0
Bridge Traffic0
0
0
0
0
0


Let’s start with the Radeons. It appears that a relatively high de-noise value helps reduce compression artifacts, something that allows even the low-end Radeons to achieve decent scores here. Surprisingly, the de-blocking option doesn’t seem to make much difference, though. Once we reach the Radeon HD 5750, the GPU is powerful enough to enable mosquito noise reduction, and this does appear to have a positive (but relatively subtle) effect on compression artifacts. We do judge that this effect is enough to earn a score of five, however, and this sets the bar for the consumer graphics cards that we’ve tested.

The GeForce cards don’t have a dedicated option to reduce compression artifacts, such as mosquito noise reduction or de-blocking, and the noise reduction setting in the Nvidia driver doesn’t appear to affect anything but grainy noise. Because we can’t see any difference with noise reduction on or off, we’re forced to give the GeForce cards a zero score in this test.

Chapter 3: Upscaled Compression Artifacts Tests

The tests here are identical to the ones we looked at in Chapter 2, the only difference being that the video is now upscaled from a lower resolution. This results in more dramatic compression artifacts and looks more like something you might see from a low-resolution source—think YouTube video.

The GeForce cards still have a hard time with compression artifacts, and the Radeons struggle as well here. As a result, AMD's cards do not achieve the high scores across the board that they did in the previous test.

Upscaled Compression Artifacts Test Results (out of 5)

Radeon HD 6850Radeon HD 5750Radeon HD 5670Radeon HD 5550Radeon HD 5450
Scrolling Text
3
3
3
3
3
Roller Coaster
5
5
3
3
3
Ferris Wheel
3
3
3
3
3
Bridge Traffic
3
3
3
3
3

GeForce GTX 470GeForce GTX 460GeForce 9800 GTGeForce GT 240GeForce GT 430GeForce 210
Scrolling Text0
0
0
0
0
0
Roller Coaster0
0
0
0
0
0
Ferris Wheel0
0
0
0
0
0
Bridge Traffic0
0
0
0
0
0
Ask a Category Expert

Create a new thread in the Reviews comments forum about this subject

Example: Notebook, Android, SSD hard drive

Display all 84 comments.
This thread is closed for comments
Top Comments
  • 14 Hide
    rootheday , February 2, 2011 3:18 AM
    Could you give the same evaluation to Sandy Bridge's Intel HD Graphics?.
Other Comments
  • 14 Hide
    rootheday , February 2, 2011 3:18 AM
    Could you give the same evaluation to Sandy Bridge's Intel HD Graphics?.
  • 4 Hide
    jimmysmitty , February 2, 2011 3:44 AM
    I second the test using SB HD graphics. It might be just an IGP but I would like to see the quality in case I want to make a HTPC and since SB has amazing encoding/decoding results compared to anything else out there (even $500+ GPUs) it would be nice to see if it can give decent picture quality.

    But as for the results, I am not that suprised. Even when their GPUs might not perform the same as nVidia, ATI has always had great image quality enhancements, even before CCC. Thats an area of focus that nVidia might not see as important when it is. I want my Blu-Ray and DVDs to look great, not just ok.
  • 1 Hide
    compton , February 2, 2011 4:10 AM
    Great article. I had wondered what the testing criteria was about, and Lo! Tom's to the rescue. I have 4 primary devices that I use to watch Netflix's streaming service. Each is radically different in terms of hardware. They all look pretty good. But they all work differently. Using my 47" LG LED TV I did an informal comparison of each.

    My desktop, which uses a 460 really suffers from the lack of noise reduction options.
    My Samsung BD player looks less spectacular that the others.
    My Xbox looks a little better than the BD player.
    My PS3 actually looks the best to me, no matter what display I use.

    I'm not sure why, but it's the only one I could pick out just based on it's image quality. Netflix streaming is basically all I use my PS3 for. Compared to it, my desktop looks good and has several options to tweak but doesn't come close. I don't know how the PS3 stacks up, but I'm thinking about giving the test suite a spin.

    Thanks for the awesome article.
  • 4 Hide
    cleeve , February 2, 2011 4:17 AM
    Quote:
    Could you give the same evaluation to Sandy Bridge's Intel HD Graphics?.


    That's Definitely on our to-do list!

    Trying to organize that one now.
  • 1 Hide
    lucuis , February 2, 2011 4:58 AM
    Too bad this stuff usually makes things look worse. I tried out the full array of settings on my GTX 470 in multiple BD Rips of varying quality, most very good.

    Noise reduction did next to nothing. And in many cases causes blockiness.

    Dynamic Contrast in many cases does make things look better, but in some it revealed tons of noise in the greyscale which the noise reduction doesn't remove...not even a little.

    Color correction seemed to make anything blueish bluer, even purples.

    Edge correction seems to sharpen some details, but introduces noise after about 20%.

    All in all, bunch of worthless settings.
  • 0 Hide
    killerclick , February 2, 2011 5:13 AM
    jimmysmittyEven when their GPUs might not perform the same as nVidia, ATI has always had great image quality enhancements


    ATI/AMD is demolishing nVidia in all price segments on performance and power efficiency... and image quality.
  • -2 Hide
    alidan , February 2, 2011 5:34 AM
    killerclickATI/AMD is demolishing nVidia in all price segments on performance and power efficiency... and image quality.


    i thought they were loosing, not by enough to call it a loss, but not as good and the latest nvidia refreshes. but i got a 5770 due to its power consumption, i didn't have to swap out my psu to put it in and that was the deciding factor for me.
  • 3 Hide
    haplo602 , February 2, 2011 5:40 AM
    this made me lol ...

    1. cadence tests ... why do you marginalise the 2:2 cadence ? these cards are not US exclusive. The rest of the world has the same requirements for picture quality.

    2. skin tone correction: I see this as an error on the part of the card to even include this. why are you correcting something that the video creator wanted to be as it is ? I mean the movie is checked by video profesionals for anything they don't want there. not completely correct skin tones are part of the product by design. this test should not even exist.

    3. dynamic contrast: cannot help it, but the example scene with the cats had blown higlights on my laptopt LCD in the "correct" part. how can you judge that if the constraint is the display device and not the GPU itself ? after all you can output on a 6-bit LCD or on a 10-bit LCD. the card does not have to know that ...
  • 4 Hide
    mitch074 , February 2, 2011 5:43 AM
    "obscure" cadence detection? Oh, of course... Nevermind that a few countries do use PAL and its 50Hz cadence on movies, and that it's frustrating to those few people who watch movies outside of the Pacific zone... As in, Europe, Africa, and parts of Asia up to and including mainland China.

    It's only worth more than half the world population, after all.
  • 1 Hide
    cleeve , February 2, 2011 5:54 AM
    mitch074"obscure" cadence detection? Oh, of course... Nevermind that a few countries do use PAL and its 50Hz cadence on movies...


    You misunderstand the text, I think.

    To clear it up: I wasn't talking about 2:2 when I said that, I was talking about the Multi-Cadence Tests: 8 FPS animation, etc.
  • 0 Hide
    caeden , February 2, 2011 5:58 AM
    Now I am not sure how the radions stack up, but using these settings on my 9800GT made my video quality noticeably worse than normal. Especially around text (white over black background was painful) where noise was added. After messing with it a little I found that using the video defaults under the image settings was best. And checking the dynamic contrast under color settings helped the contrast a little, but not much, and I think it is adding noise, so I am going back to defaults on this as well.

    More important than your driver would be calibrating your monitor. PC monitors are very different than HDTVs (higher brightness and lower contrast, square pix to optimize text over video, and most of us have tn screens which are very limited on color range), but in spite of their shortcomings, if properly adjusted they can still look quite good.

    A note to those who are putting your DVDs on your computer, there is a very powerful program called AVS which can do much of this cleanup for you during the ripping process. This allows for smaller file size with good compressors (h264 for storage, Lagarith for editing), as well as making upscaling less of a problem. It is not a particularly intuitive tool, but there are many good tutorials out there on it. This is especially handy with old DVDs that were not encoded well in the first place, or super long DVDs (lord of the rings) where it was encoded poorly in order to fit the content on the disc.
    As for blue-Ray content, the quality you get will depend mostly on the software you are using. The driver options (at least for my card) hinder more than help, so it then depends on the software you are using.
    For HD content in general you will notice a huge difference between that nicely encoded music video you download vs Netflix HD streaming. Part of this is because Netflix is (for the most part) only 720p, and 2ndly because they have an automated process which will work better for some movies than others. It would seem their biggest flaw is color banding. This is still worlds better than cable, and other web services, but you are just not going to get that BD quality.
  • 1 Hide
    cleeve , February 2, 2011 6:03 AM
    haplo602 cadence tests ... why do you marginalise the 2:2 cadence ? these cards are not US exclusive. The rest of the world has the same requirements for picture quality.


    True enough, although I'm a US writer primarily tasked to write for the US audience. I do think readers from PAL countries can understand the implications, however.

    haplo602 2. skin tone correction: I see this as an error on the part of the card to even include this. why are you correcting something that the video creator wanted to be as it is ? I mean the movie is checked by video profesionals for anything they don't want there. not completely correct skin tones are part of the product by design. this test should not even exist.


    I understand the argument, and frankly the fact that user is free to turn it off is good enough for me. Regardless of whether or not we agree with its inclusion, it is something that's included in some high-end video processors and is therefore a point of comparison.

    haplo602 3. dynamic contrast: cannot help it, but the example scene with the cats had blown higlights on my laptopt LCD in the "correct" part. how can you judge that if the constraint is the display device and not the GPU itself ? after all you can output on a 6-bit LCD or on a 10-bit LCD. the card does not have to know that ...


    For me it was sufficient to use a good-quality display for testing the different graphics hardware. Frankly, the HQV test can be used for displays, too. But with our test display capable of differentiating contrast enhancement without overexposure I think that's the best way to test the video processor. Whether other folks have worse displays is kind of outside the scope of the test.
  • 3 Hide
    spac18 , February 2, 2011 6:42 AM
    @shinobi
    where did you get that stat from? your all american wet dream?
  • 3 Hide
    intelx , February 2, 2011 6:47 AM
    shin0bi272but 85% of the worlds computers are in the US and so your countries dont really matter.


    can you prove your statement?
  • 4 Hide
    andrewcutter , February 2, 2011 7:45 AM
    shin0bi272but 85% of the worlds computers are in the US and so your countries dont really matter.


    why am i not suprised by this statement. i hope you are not representative of your whole country
  • 4 Hide
    aldaia , February 2, 2011 8:46 AM
    shin0bi272but 85% of the worlds computers are in the US and so your countries dont really matter.


    I'm not sure if he is a troll or simply a dumb ass (probably both). The opposite is actually closer to reality, that is 85% of personal computers are outside of the US. In 2005-2004 only 28.4% of the world computers where in the US. Since then, countries like China or India have grown significantly faster than western countries where the market was already saturated 5 years ago. By the way the countries with more computers (per capita) are Switzerland: 864.584 per 1 million people, San Marino: 857.143 per 1 million people & Sweden: 763.012 per 1 million people.
  • 3 Hide
    haplo602 , February 2, 2011 9:32 AM
    CleeveTrue enough, although I'm a US writer primarily tasked to write for the US audience. I do think readers from PAL countries can understand the implications, however.


    I do not dispute the results as they speak clearly. I do have a problem with you downplaying an Nvidia disadvantage just because you are a US resident. The tone of the comment was not neutral (i.e. just to highlight that one cadence is used for 60Hz and the other for 50Hz).

    Nvidia does not issue non-US driver versions or non-US cards, so they should be blamed for the lack of features in this case.

    I just want to make you aware that some of your comments might not come across as correct to non-US residents. After all, Internet knows no borders :-)
  • -1 Hide
    marraco , February 2, 2011 9:52 AM
    It's useless without a video player capable of applying all those enhancements to different video formats.
  • 2 Hide
    hardcore_gamer , February 2, 2011 10:24 AM
    CleeveThat's Definitely on our to-do list!Trying to organize that one now.

    Please add gtx5xx series cards too
Display more comments