Neal1981
19th January 2011, 09:42
Dear all,
Recently, I studied the code of x264 and find there's something different as I know to the quantization part.
From SPEC, it had the similar formula.
Coeff = (Coeff*Quantizer+QP_Const)>>Q_Bits
where, Q_Bits = QP/6 + 15,
QP_Const = (1<<Q_Bits)/3 for intra,
QP_Const = (1<<Q_Bits)/6 for inter
However, in x264, it fixed the Q_Bits to always 16, and scale "(Coeff*Quantizer+QP_Const)" by some specific bias factor.
Does anyone know the purpose of this design? or is any paper can help to reference?
Deeply appreciate for the feedback.
Recently, I studied the code of x264 and find there's something different as I know to the quantization part.
From SPEC, it had the similar formula.
Coeff = (Coeff*Quantizer+QP_Const)>>Q_Bits
where, Q_Bits = QP/6 + 15,
QP_Const = (1<<Q_Bits)/3 for intra,
QP_Const = (1<<Q_Bits)/6 for inter
However, in x264, it fixed the Q_Bits to always 16, and scale "(Coeff*Quantizer+QP_Const)" by some specific bias factor.
Does anyone know the purpose of this design? or is any paper can help to reference?
Deeply appreciate for the feedback.