Is JPEG compression considered a fast algorithm? Are there faster/slower methods of compression that come with different tradeoffs?
stephen422
Wouldn't the "compute DCT" step also introduce loss? Because we only use 64 bases for the transform, and it projects the infinite-dimensional space of the original signal to a 64-dimensional space.
kevintli
I think the reason the DCT step doesn't result in any error is because it's converting individual 8x8 pixel blocks into frequency space; since pixels are already a discretized version of the original signal, we aren't actually working in an infinite-dimensional space of frequencies, and thus all possible frequencies can be represented by the 64 bases.
Is JPEG compression considered a fast algorithm? Are there faster/slower methods of compression that come with different tradeoffs?
Wouldn't the "compute DCT" step also introduce loss? Because we only use 64 bases for the transform, and it projects the infinite-dimensional space of the original signal to a 64-dimensional space.
I think the reason the DCT step doesn't result in any error is because it's converting individual 8x8 pixel blocks into frequency space; since pixels are already a discretized version of the original signal, we aren't actually working in an infinite-dimensional space of frequencies, and thus all possible frequencies can be represented by the 64 bases.