Sampling

Latest revision as of 21:38, 15 June 2016

[edit] Color format conversions and the Chroma Upsampling Error

The following figures show errors, which are examples of the Chroma Upsampling Error, called this way because the video is upsampled incorrectly (interlaced YV12 upsampled as progressive or vice versa). As a result, you will often see gaps on the top and bottom of colored objects and "ghost" lines floating above or below the objects.

figure 1a: example of interlaced source (YV12) being upsampled as progressive video (YUY2)

figure 1b: the same image with correct chroma upsampling

figure 2a: example of progressive source (YV12) being upsampled as interlaced video (YUY2)

figure 2b: the same image with correct chroma upsampling

In this section, it will be shown what causes it, and how to fix it. Where fixing mean making it less visible, because it's not possible to correct it completely.

References:

• The 4:2:0 Interlaced Chroma Problem
• The 4:2:0 Interlaced Chroma Problem - Television and Video Advice

[edit] Chroma Upsampling Error (or CUE)

As previously stated, the Chroma Upsampling Error occurs when you convert from (truly) interlaced YV12 to mostly any other format and the converter thinks the video is progressive. Or, the other way around, if material is progressive (or interlaced and encoded as progressive), and upsampled as interlaced. This is however not as bad as the other way around.

When Virtualdub previews your video, it will need to convert it to RGB. Since AviSynth delivers YV12, it asks the codec (for example XviD or DivX) to convert YV12 to RGB. The codec however ALWAYS upsamples progressively. Hence you will get artifacts in VDub preview on interlaced YV12 material. This is however not present in the YV12 video (or in the resulting encoding). To confirm this, let AviSynth do the conversion by adding ConvertToRGB(interlaced=true) at the end of your script.

[edit] Correcting video having the Chroma Upsampling Error

You will have to blur the chroma in some way (leaving the luma intact).

For example (using TomsMoComp):

AviSource(...)
MergeChroma(TomsMoComp(-1,5,0))

[edit] Theoretical Aspects

In this section the chroma placement will be explained, how this is related to subsampling (RGB -> YUY2 -> YV12) and how the upsampling is done in AviSynth.

It should also explain in detail why the CUE occurs. To summarize the latter, the problem is that there is a difference between YV12 progressive and YV12 interlaced, because the chroma is shared vertically between neighboring pixels.

See also this doom9 thread.

[edit] The color formats: RGB, YUY2 and YV12

In order to be able to understand how YV12 <-> YUY2 sampling works and why it matters whether your source is interlaced or progressive, the YV12/YUY2 color formats will be discussed first. It's not important here how they are stored in your memory. Information about that can be found here: Color Spaces.

[edit] YUV 4:4:4 color format

The term 4:4:4 denotes that for every four samples of the luminance (Y), there are four samples each of U and V. Thus each pixel has a luminance value (Y), a U value (blue difference sample or Cb) and a V value (red difference sample or Cr). Note, "C" is just a chroma sample (UV-sample).

The layout of a 4:4:4 encoded image looks as follows

[edit] YUY2 color format

YUY2 (or YUYV) is a 4:2:2 format. The term 4:2:2 denotes that for every four samples of the luminance (Y), there are two samples each of U and V, giving less chrominance (color) bandwidth in relation to luminance. So for each pixel, it is horizontally sharing UV (chroma) with a neighboring pixel.

The layout of a 4:2:2 encoded image looks as follows

[edit] YV12 color format

For the YV12 color format, there's a difference between progressive and interlaced. The cause is that chroma values are also shared vertically between two neighboring lines.

YV12 is a 4:2:0 format. The term 4:2:0 denotes that for every four samples (two horizontal and two vertical) of the luminance (Y), there is one sample each of U and V, giving less chrominance (color) bandwidth in relation to luminance.

YV12 progressive

For each pixel, it is horizontally sharing UV (chroma or C) with a neighboring pixel and vertically sharing UV with the neighboring line (thus line 1 with line 2, line 3 with 4, etc).

The layout of a progressive 4:2:0 encoded image looks as follows (MPEG 2 scheme - see below)

YV12 interlaced

For each pixel, it is horizontally sharing UV (chroma or C) with a neighboring pixel and vertically sharing UV with the next to neighboring line (thus line 1t with line 3t, line 2b with 4b, etc).

The layout of a interlaced 4:2:0 encoded image looks as follows (MPEG 2 scheme - see below)

or

[edit] Subsampling

Subsampling is used to reduce the storage and broadcast bandwidth requirements for digital video. This is effective for a YCbCr signal because the human eye is more sensitive for changes in black and white than for changes in color. So drastically reducing the color info shows very little difference. YUY2 and YV12 are examples of reduced color formats.

[edit] RGB -> YUY2 conversion

More about RGB -> YUV color conversions can be found here: Color Conversions. Recall the layout of a 4:4:4 encoded image

In AviSynth, the default mode is using a [1 2 1] kernel to interpolate chroma, that is

C1x = (2*C1+C2)/3 (I'm not sure how the edge pixel gets interpolated)
C3x = (C2+2*C3+C4)/4
C5x = (C4+2*C5+C6)/4

The 4:2:2 encoded image becomes

(In v2.58, the [0 1 1] kernel was used to interpolate chroma. This is technically wrong and results in a 0.5 pixel right shift of the chroma, but it was originally done for performance reasons.)

The other mode ConvertBackToYUY2 uses the chroma from the left pixel (so the chroma is point-sampled), thus

Note (as with the layout of other formats) the position of the chroma values, represent the WEIGHT result of the subsampling.

[edit] YUY2 -> YV12 interlaced conversion

Recall the layout of a interlaced 4:2:0 encoded image, but with the weights included:

or

Note (as with the layout of other formats) the position of the chroma values, represent the WEIGHT as a result of the subsampling.

Thus the chroma is stretched across two luma lines in the same field!

[edit] YUY2 -> YV12 progressive conversion

Recall the layout of a 4:2:0 encoded image

Note (as with the layout of other formats) the position of the chroma values, represent the WEIGHT result of the subsampling.

Thus the chroma is stretched across two luma lines in the same frame!

[edit] Upsampling

[edit] YUY2 conversion -> RGB

Recall the layout of a 4:2:2 encoded image

The C++ and the ASM code use different sampling methods.

For the C++ code ConvertToRGB uses the same chroma for two RGB pixels using left point upsampling. Thus the 4:4:4 encoded image becomes

For the ASM code ConvertToRGB the missing chroma samples are interpolated (using a [1 1] kernel), that is

C2x = (C1+C3)/2
C4x = (C3+C5)/2

and the existing chroma samples are just copied.

Thus the 4:4:4 encoded image becomes

The ASM code is the one which is actually used in AviSynth.

[edit] YV12 interlaced conversion -> YUY2

In AviSynth, the missing chroma samples are interpolated as follows

or

This implementation results in a chroma shift. AviSynth uses a different interpolation as the one suggested by the mpeg-2 specs (perhaps due to speed issues). The latter is

[edit] YV12 progressive conversion -> YUY2

The missing chroma samples are interpolated as follows

[edit] References

ColorSpaces

• 4:4:4 sampling
• 4:2:2 sampling
• 4:2:0 sampling
• 4:1:1 sampling
• Chroma Upsampling
• Chroma Subsampling Standards

[edit] mpeg-1 versus mpeg-2 sampling

There are two common variants of 4:2:0 sampling. One of these is used in MPEG-2 (and CCIR-601) video, and the other is used in MPEG-1. The MPEG-2 scheme is how AviSynth samples 4:2:0 video, because it completely avoids horizontal resampling in 4:2:0 <-> 4:2:2 conversions.

The layout of a progressive MPEG-1 4:2:0 encoded image

The layout of a MPEG-2 4:2:0 encoded image

[edit] DV sampling

For completeness, we will mention DV sampling. DV is 4:2:0 (PAL) and 4:1:1 (NTSC). Note, that the sample positioning of the former is different from the 4:2:0 chroma in MPEG-1/MPEG-2!

The layout of a 4:2:0 encoded image (field-based)

The layout of a 4:1:1 encoded image

Some comments about this formats:

• 4:1:1 is supported natively in AviSynth v2.6.
• DV decoders all output YUY2 or RGB (with the exception of ffdshow when YV12 is enabled).
• When outputting YUY2/RGB (NTSC), the MainConcept codec duplicates the chroma samples instead of interpolating. The ReInterpolate411 plugin can be used to correct for this, resulting in better quality.

[edit] References

• MSDN: YUV sampling Describes the most common YUV sampling techniques.
• DV sampling

[edit] 4:2:0 Interlaced Chroma Problem (or ICP)

In general interlaced content will have static parts. If it is upsampled correctly using interlaced upsampling, it will still have chroma problems on diagonal edges of bright-colored objects in static parts of a frame.

The reason is that "When the two fields are put back together later by a deinterlacer (or by your eye and brain, if you watch it on an interlaced TV), the relatively smooth gradations and contours of each field are broken up by a slightly different set of gradations and contours from the other field." (quote from first reference). This is called the Interlaced Chroma Problem.

The "solution" is a motion-adaptive upsampler. The only known native Avisynth filter performing this function is available at http://neuron2.net/autoyuy2/autoyuy2.html, although a slower Avisynth script has also been implemented.

References:

• The 4:2:0 Interlaced Chroma Problem
• The 4:2:0 Interlaced Chroma Problem - Television and Video Advice