3 ms·
It enables applications with spectral frequency components that are not constrained to have fixed magnitudes but also could decay or grow exponentially. I th
by threeme3 7y ago
It enables applications
with spectral frequency components that are not constrained to have fixed magnitudes but also could decay or
grow exponentially.
I think this exponential decay or growth is the major difference compared to an IFFT, and hence the algorithm can work more efficient over a larger time/frequency domain.
Think of improvements anywhere FFT is used today, but where there is a need to work over a non-linear time/frequency domain.
- taneq 7y agoMy layman’s interpretation of this is that using this new transform instead of DCT will make edges less blocky in JPG images. Is that right?
- mkl 7y agoNot really. The blockyness in JPEG images is from throwing away less significant terms in each 8x8 pixel block to save space. Any lossy image compression method will cause artefacts of some kind, no matter what transform or system it's based on, but JPEG's ones are particularly artificial-looking as they're square and lead to sharp visible edges. Some other methods of throwing away information aren't based on blocks and just make things blurry.
- magicalhippo 7y agoAFAIK the main source of blockyness is due to the quantization of low-frequency components, such that the decoded signal doesn't quite match from one block to the next. The ringing around edges in the source signal, Gibbs phenomenon[1], is due to almost or entirely discarding higher frequency components. [1]: https://en.wikipedia.org/wiki/Gibbs_phenomenon https://en.wikipedia.org/wiki/Gibbs_phenomenon