| dc.description.abstract | In today′s society, our demand for resolution is increasing. To meet the requirements for high-resolution images, Versatile Video Coding (VVC) can achieve a compression rate twice as high as the previous generation video coding standard, High Efficiency Video Coding (HEVC). This improvement is attributed to the use of coding units, prediction units, transform units, and quantization in HEVC′s compression techniques.
In terms of network transmission, bitrate control aims to achieve lower distortion and better performance of transmitted images within a specific channel capacity. This paper adopts an R-λ model to control the bit rate at the Coding Tree Unit (CTU) level under low-delay configuration. For inter-frame prediction, it is observed that the bit error rate of some CTUs is positively correlated with pixel domain variance and motion vector variance. The bitrate control algorithm under the R-λ model heavily relies on the accuracy of the λ value to achieve precise target bit rates.
In this paper, for more accurate bitrate control in inter-frame prediction, a statistical approach is used to determine specific thresholds for pixel domain variance and motion vector variance. If these variances exceed the threshold, a Convolutional Neural Network (CNN) is invoked to predict the parameters of the coding tree unit. Experimental results demonstrate that the CNN-based approach, compared to the bitrate control method in VTM 20.0 of VVC, reduces the bit error rate by 0.122%. | en_US |