Hi there,
Thanks for checking out a 'Explained' blog! In these; my aim is to give you some details on what some common technical terms mean in digital video and film. I hope you find something useful here.
Please note; I do not consider myself an expert on the topics - if you believe I've made a mistake somewhere please do let us know! I'd love for this to be a learning experience for me as much as anyone else.
Chroma Subsampling
In this blog we're going to look at Chroma Subsampling. It is definitely worth checking out the Colour Models & Spaces blog first if you haven’t already. This is a fairly complex area, and as mentioned, please don’t regard what you read here as fact; this is merely my understanding of it as I can gain from my independent research; it is not a comprehensive breakdown, more an introduction into the topic.
Chroma-Subsampling is a technique used in video compression that reduces the file size and therefore bandwidth requirements of video files by removing colour information. It is made possible by using the YCbCr Colour Model.
You may have seen values such as 4:4:4, 4:2:2, 4:2:0 etc on camera specifications, video files and elsewhere; these represent the type of chroma-subsampling being used in the compression, but how exactly does Chroma-subsampling work?
The principle behind it is based on the idea that the human eye is much more sensitive to variations in brightness than it is to colour. This is why YCbCr separates the luminance channel from the colour ones, so that chroma subsampling can take advantage of this knowledge to reduce the data of the colour information whilst preserving that of the brightness. The best way to understand how this works is through diagrams (shout out to Karl Soule, who's post on this I find to be most clear and understandable; and is also the source of these images):
You may have seen values such as 4:4:4, 4:2:2, 4:2:0 etc on camera specifications, video files and elsewhere; these represent the type of chroma-subsampling being used in the compression, but how exactly does Chroma-subsampling work?
The principle behind it is based on the idea that the human eye is much more sensitive to variations in brightness than it is to colour. This is why YCbCr separates the luminance channel from the colour ones, so that chroma subsampling can take advantage of this knowledge to reduce the data of the colour information whilst preserving that of the brightness. The best way to understand how this works is through diagrams (shout out to Karl Soule, who's post on this I find to be most clear and understandable; and is also the source of these images):
![Picture](/uploads/2/4/5/7/24574909/1758040.gif)
4:4:4
The diagram to the left shows us a 4x4 grid, where each square represents a pixel. Imagine this grid is just a cross-section of a full video frame. Within each pixel we have 3 squares, black, red and blue. Each of these are a separate channel, the three of which make up the final pixel value. Black is luminance, red is difference-red, and blue is, you guessed it, difference-blue.
As you can tell, 4:4:4 is more or less what you would expect, each pixel contains information for every channel in it, no subsampling is taking place in this instance. Unless you work with high quality video, you're unlikely to encounter 4:4:4 footage often - it's expensive to achieve/work with and is generally unnecessary.
The diagram to the left shows us a 4x4 grid, where each square represents a pixel. Imagine this grid is just a cross-section of a full video frame. Within each pixel we have 3 squares, black, red and blue. Each of these are a separate channel, the three of which make up the final pixel value. Black is luminance, red is difference-red, and blue is, you guessed it, difference-blue.
As you can tell, 4:4:4 is more or less what you would expect, each pixel contains information for every channel in it, no subsampling is taking place in this instance. Unless you work with high quality video, you're unlikely to encounter 4:4:4 footage often - it's expensive to achieve/work with and is generally unnecessary.
![Picture](/uploads/2/4/5/7/24574909/9406975.gif)
4:2:2
As you can see, we've completely cut out the colour information from every other column of pixels. This may seem like a lot of missing data, but this is still high quality. There is currently no digital video distribution platform that handles 4:2:2, it's only really used during the production process.
So how is it we are completely oblivious to the removal of this information? You'd be hard up to see a difference even between side by side comparisons, particularly 4:4:4 and 4:2:2.
Well, firstly it comes back to the fact that our eyes can't make out colour that well, we just don't notice it. But that's not all, when we watch subsampled footage, pixels aren't just missing colour, they all have colour in fact, it's just that the decoding software is approximating what those pixels without colour values may be based on those around it, and decoders nowadays can do a pretty good job of this. Yes, that does mean there are inaccurate reproductions, but it doesn't matter, we wouldn't notice.
As you can see, we've completely cut out the colour information from every other column of pixels. This may seem like a lot of missing data, but this is still high quality. There is currently no digital video distribution platform that handles 4:2:2, it's only really used during the production process.
So how is it we are completely oblivious to the removal of this information? You'd be hard up to see a difference even between side by side comparisons, particularly 4:4:4 and 4:2:2.
Well, firstly it comes back to the fact that our eyes can't make out colour that well, we just don't notice it. But that's not all, when we watch subsampled footage, pixels aren't just missing colour, they all have colour in fact, it's just that the decoding software is approximating what those pixels without colour values may be based on those around it, and decoders nowadays can do a pretty good job of this. Yes, that does mean there are inaccurate reproductions, but it doesn't matter, we wouldn't notice.
![Picture](/uploads/2/4/5/7/24574909/4888298.gif)
4:1:1
Now we are really starting to see some data loss, we've cut out 75% of the colour information leaving us with only one column of colour values per every 4 columns.
By this point we are starting to see many more inaccuracies in colour reproduction; it is still not as bad as might first think, most distribution platforms use this level of subsampling, albeit usually using the 4:2:0 model (see below). The decoders are still capable of reproducing colours accurate enough that the image appears normal.
Now we are really starting to see some data loss, we've cut out 75% of the colour information leaving us with only one column of colour values per every 4 columns.
By this point we are starting to see many more inaccuracies in colour reproduction; it is still not as bad as might first think, most distribution platforms use this level of subsampling, albeit usually using the 4:2:0 model (see below). The decoders are still capable of reproducing colours accurate enough that the image appears normal.
![Picture](/uploads/2/4/5/7/24574909/1660860.gif)
4:2:0
Though there are others out there, this is the last we are going to look at here. This is where things get a little different; 4:2:0 has the exact same factor of compression as 4:1:1, in that there is still the same amount of channels per 4x4 grid of pixels, they are just laid out differently. Instead of having one column of complete pixels, we have 2 columns of partially completed pixels. This means the system doesn't have to look as far as 4:1:1 to find colour information, but every pixel needs some amount of prediction to be taken place.
This type of subsampling has seemingly won out over 4:1:0 and is used in DTV, HDTV, DVD and Blu-ray.
Though there are others out there, this is the last we are going to look at here. This is where things get a little different; 4:2:0 has the exact same factor of compression as 4:1:1, in that there is still the same amount of channels per 4x4 grid of pixels, they are just laid out differently. Instead of having one column of complete pixels, we have 2 columns of partially completed pixels. This means the system doesn't have to look as far as 4:1:1 to find colour information, but every pixel needs some amount of prediction to be taken place.
This type of subsampling has seemingly won out over 4:1:0 and is used in DTV, HDTV, DVD and Blu-ray.
So that's a basic introduction into Chroma-Subsampling; I hope you've learnt something! Remember, 4:4:4 is the golden standard but out of reach & arguably unnecessary for most situations, 4:2:2 is a great source that is often considered 'visually lossless' (there is no noticeable degradation to naked eye) whilst 4:2:0 and 4:1:1 are more compressed but are the current standard for distribution. It's worth noting that it's more or less totally pointless converting a more heavily subsampled file to a less so one (such as 4:2:0 to 4:2:2), you'll gain nothing more than larger file sizes as it creates redundant colour information no more accurate than the source.