This noticeably happens in video streaming as well, I wonder if a lot of streaming formats uses the same compression algorithm underneath the hood as jpeg?
ncastaneda02
Streaming algorithms are a little more complicated because the artifacts (typically) don't arise from compression, they arise from not having enough data to fully reconstruct the image. A really popular streaming algorithm are fountain codes, which send snippets of data in packets that allow you to reconstruct the image eventually. If your bandwidth is really good, you can always have full quality loss-less video, but because that isn't always the case, you end up with artifacts sometimes because you aren't able to fully reconstruct the source image. This is pretty similar to SVD dimensionality reduction, where you can get a pretty good approximation of a matrix (in this case a frame buffer) with only a few packets.
joeyzhao123
I'm a bit curious here. I've seen the term lossless compression used some places. Do we actually not lose any data at all? And if so, how would that even work since we're reducing the size and not losing anything?
AlexSchedel
@joeyzhai123 I would recommend taking a look at Huffman Encoding (https://en.wikipedia.org/wiki/Huffman_coding) if you want an example of a pretty simple lossless algorithm. In short the way it works is by changing the representation of your data in binary. For example, instead of representing every char as 8 bits you could represent the more common ones with 1 - 4 bits and the less common ones with 8 - 12. This makes the least common chars more expensive to store, but if there are few of them the benefits of storing common chars more efficiently outweighs the cost.
joeyhou0804
I remember noticing a similar type of artifact in video streaming where there are block-like patterns. I remember, on the other hand, that this type of artifact is different from this compression artifacts.
This noticeably happens in video streaming as well, I wonder if a lot of streaming formats uses the same compression algorithm underneath the hood as jpeg?
Streaming algorithms are a little more complicated because the artifacts (typically) don't arise from compression, they arise from not having enough data to fully reconstruct the image. A really popular streaming algorithm are fountain codes, which send snippets of data in packets that allow you to reconstruct the image eventually. If your bandwidth is really good, you can always have full quality loss-less video, but because that isn't always the case, you end up with artifacts sometimes because you aren't able to fully reconstruct the source image. This is pretty similar to SVD dimensionality reduction, where you can get a pretty good approximation of a matrix (in this case a frame buffer) with only a few packets.
I'm a bit curious here. I've seen the term lossless compression used some places. Do we actually not lose any data at all? And if so, how would that even work since we're reducing the size and not losing anything?
@joeyzhai123 I would recommend taking a look at Huffman Encoding (https://en.wikipedia.org/wiki/Huffman_coding) if you want an example of a pretty simple lossless algorithm. In short the way it works is by changing the representation of your data in binary. For example, instead of representing every char as 8 bits you could represent the more common ones with 1 - 4 bits and the less common ones with 8 - 12. This makes the least common chars more expensive to store, but if there are few of them the benefits of storing common chars more efficiently outweighs the cost.
I remember noticing a similar type of artifact in video streaming where there are block-like patterns. I remember, on the other hand, that this type of artifact is different from this compression artifacts.