Efficient Video Coding At Low Bit Rate And Video Quality Enhancement Research

Video coding technology is an effective method to reduce the amount of video data,however,in the case of low coding bit rate,the subjective and objective quality of video is often poor.The experiments in this thesis show that in the case of limited bandwidth,there are often problems such as smearing,large-area blur and large-area block effect in low-bit-rate encoded videos.Therefore,this thesis focuses on the subject of high-efficiency video coding and video quality enhancement at low bit rates,and uses deep learning algorithms with outstanding performance in some image and video tasks and variable-resolution algorithms based on high-efficiency video coding standards to conduct in-depth exploration.1.For the high-efficiency video coding scheme at low bit rate,this thesis proposes two variable resolution algorithm schemes in which the variational frame is I frame and P frame.Experiments show that the two schemes have similar effects on image quality improvement.The variable resolution algorithm scheme based on I frame is an algorithm scheme that complies with the coding standard,but has the defect of rate expansion.The variable resolution algorithm scheme based on P frame is an algorithm scheme that does not meet the coding standard.In this thesis,it is proved by experiments that in the case of extremely low bit rate,the use of variational coding algorithm can greatly enhance the quality of video frames,and the objective PSNR performance index can achieve up to2 d B improvement.In terms of bit rate savings,using the variable resolution algorithm can save 10% to 47% of the bit rate under the same image quality.Based on the idea of using the entire video communication link to improve the quality of the encoded video,this thesis adopts the post-processing technology of denoising and sharpening to make up for the blurring of details caused by the variable resolution algorithm,which complements its advantages and further improves the final video quality.Finally,the work of this thesis explores the scope of application of variable resolution technology.The conclusion is that this technology is often suitable for low bit rate segments,and the effect is weakened as the bit rate increases.2.For the compressed video quality enhancement scheme based on deep learning,this thesis proposes a deep learning network design idea based on traditional algorithms,and on this basis,designs a network model for feature domain alignment,residual network information reconstruction and a GAN-based network model which is an algorithmic models for adversarial thinking.The experimental results show that the network architecture achieves a good deblocking and deblurring effect on the tested video sequences,and the objective PSNR performance index achieves an improvement of0.1?0.4d B.