i shortly outline an exponential frame-level Rate Control model, where QP (quantization parameter) is constant (or near-constant) within a single frame. Notice that MB-level rate control can achieve more accurate coded bitrate than frame-level because the quantization parameter (QP) may be changed with different MBs.
Exponential Adaptive Rate Control is based on exponential R-Q model:
where
R – estimated frame size
X – complexity measure
QP – quantization parameter (may be replaced with Qstep ) applied to all MBs in a frame
α>0 and β<0 are model parameters.
Notice that the parameter β can be chosen β=-1 as an approximation of memoryless Gaussian Rate Distortion R-D curve:
One of complexity metrics is G-metric: the average gradient per pixel metric computed for NxM image:
Notice that the complexity G can be computed for each frame in pre-processing phase.
To make the model adaptive the parameter α is updated after encoding of each frame (e.g. k-th frame) as follows:
Here
Xk – k-th frame complexity
Ract(k) – actual bit-size of the k-th frame
QPk – k-th frame quantization parameter
θ – adaptation resistance factor (commonly taken as 0.5)
Thus, α is only adaptive parameter, β is pre-defined and fixed.
23+ years’ programming and theoretical experience in the computer science fields such as video compression, media streaming and artificial intelligence (co-author of several papers and patents).
the author is looking for new job, my resume
