Research On Rate Control Technology In X264 For Conference Video

There is a growing demand for video conferencing and online classes,affected by COVID-19.The current desktop applications such as enterprise video conferencing and university cloud classes have exploded.How to ensure both the quality of the picture and the low latency feeling of users is the biggest challenge of the current conference software system.Limited by the hardware of computer and mobile client,the encoder with high complexity is difficult to be applied.X264 encoder,as a simplified version of H.264 standard,has low requirements for hardware,and can achieve real-time coding under the condition of ensuring video quality.It is the most widely used encoder in various conference software at present.This thesis focuses on the bit rate control of X264 encoder in conference video scene,and improves the related techniques at frame level and macroblock level respectively according to the video characteristics to improve the bit rate control accuracy and coding performance.The thesis consists following research work:1.The X264 encoder simplifies the code rate control algorithm recommended by the H.264 standard,and uses average target rate ABR mode.This algorithm has poor rate control accuracy and also needs to calculate the complexity of each frame,which consumes more coding time.This thesis proposed a frame-level code rate control algorithm based on the R-lambda model.Set sliding windows to dynamically adjust the target bit allocation of encoded frames.The QP lambda curve of x264 encoder is re fitted,and the QP lambda formula in the model is further modified.At the same time,the quantization parameters of the encoded frame are cached by using the sliding window,and the variation amplitude of the quantization parameters of the current encoded frame is smoothed through its mean value,so as to eliminate the compression artifact of the reconstructed video.Compared with the ABR algorithm of X264 encoder,this method achieves better bit rate control accuracy and better coding quality with reduced computational complexity.2.If only the frame level rate control algorithm is used,the macroblocks in the same coding frame use unified frame level coding parameters,and the difference of macroblock importance is ignored.Based on the time-domain rate distortion optimization theory,this thesis calculates the time-domain influence factor of each macroblock by analyzing the time-domain dependence of the encoded macroblocks,assigns distortion weights to different macroblocks within the same frame,and thus adjusts the Lagrange multipliers of macroblocks based on the obtained frame-level Lagrange multipliers according to the distortion weights,and each macroblock is able to adopt different quantization parameters according to the importance.Compared with the ABR algorithm of X264 encoder,this method achieves better encoding performance.