Module dsp::dct

Discrete Cosine Transform (DCT-II / DCT-III)

dct2 is the "DCT-II" (what's usually just called "the DCT" — JPEG/MPEG use it for the same reason: real-valued input, real-valued output, and its energy-compaction property concentrates most of a typical signal's energy into the first few coefficients). idct3 is its exact inverse (the "DCT-III", sometimes called "the IDCT") — dct2 then idct3 reconstructs the original signal exactly (to floating-point/fixed-point rounding), the same round-trip relationship dft.h/idft and fft/ifft have.

Direct O(n^2) summation, like dft.h — see fft.h if n is a power of two and O(n log n) matters more than simplicity (there's no fast-DCT equivalent here yet; add one via an FFT-based algorithm if that becomes the bottleneck).