Adaptive Piecewise Linear Bits Estimation Model for MPEG Based Video Coding

In many video compression applications, it is essential to control precisely the bit rate produced by the encoder. One critical element in a bits/buffer control algorithm is the bits model that predicts the number of compressed bits when a certain quantization stepsize is used. In this paper, we propose an adaptive piecewise linear bits estimation model with a tree structure. Each node in the tree is associated with a linear relationship between the compressed bits and the activity measure divided by stepsize. The parameters in this relationship are adjusted by the least mean squares algorithm. The effectiveness of this algorithm is demonstrated by an example of digital VCR application. Simulation results indicate that this bits model has a fast adaptation speed even during scene changes. Compared to the bits model derived from training data based on cluster analysis, the adaptive piecewise linear bits model achieves about the same high performance with a much lower complexity and high self-adaptativity. A particular advantage of a rate control scheme employing a bits model over the buffer-feedback rate control such as MPEG2 Test Model 5 is that it can control the bits of every microblock very precisely.