pub fn dct_block_unquantized(pixels: &[f64; 64], qt: &[u16; 64]) -> [f64; 64]Expand description
8×8 forward DCT + divide by QT, but do NOT round.
Returns continuous f64 values (the “unquantized” coefficients).
The rounding error for each position is:
unquantized[i] - dct_block(pixels, qt)[i] as f64
and lies in the range [-0.5, +0.5].
Used by SI-UNIWARD to exploit quantization rounding errors for more efficient embedding when the original pixels are available.