【摘要】 为解决在有限传输带宽下海量直播视频的高质量传输受限的问题，提出一种基于感兴趣区域(ROI)的码率控制方法。根据视觉注意机制对整帧图片进行三级划分处理，提取像素级感兴趣区域图，保证不同区域平滑相接；分别以帧级和最大编码单元级(LCU)的ROI在整帧图像中所占的比例作为权重对帧层和LCU层进行目标比特的分配；再通过R-λ模型获得编码参数，实现压缩编码，同时完成模型参数更新。在通用测试条件下的实验结果表明，算法提升了视频压缩比，保证了重建视频的主观质量，在整帧图像质量的PSNR相对稳定情况下，相比于HM16.7中算法，多个测试序列中ROI的平均ΔY-PSNR增加了0.75 dB,非ROI的平均ΔY-PSNR下降了0.43 dB,平均码率相对误差为0.226%。更多还原

【Abstract】 In order to solve the issue of restricted high-quality transmission of massive live video under limited transmission bandwidth, we propose a rate control method based on region of interest(ROI). The whole frame of image was processed by three-level division according to the visual attention mechanism, and the pixel-level ROI map was extracted to make different regions smoothly connected. The target bits were allocated to the frame layer and the largest coding unit(LCU) layer by taking the proportion of ROI at the frame level and the LCU level as weights respectively. We obtained the coding parameters through R-λ model, which realized the compression coding and completed the update of the model parameters at the same time. The experimental results under general test conditions reveal that the proposed method improves the compression ratio and ensures the subjective quality of the reconstructed video. Compared with the algorithm in HM16.7, the average ΔY-PSNR of ROI in multiple test sequences increases by 0.75 dB, the average ΔY-PSNR of non-ROI decreases by 0.43 dB, and the relative error of the average bit rate is 0.226%, when the PSNR of the whole frame image quality is relatively stable.更多还原