基于人眼显著性区域检测的HEVC码率控制算法  

A HEVC Bitrate Control Algorithm Based on Human Visual Saliency Detection

作　　者：亢润龙 KANG Runlong

机构地区：[1]洛阳职业技术学院信息工程学院,河南洛阳471000

出　　处：《科技创新与应用》2024年第34期11-17,共7页Technology Innovation and Application

摘　　要：该文提出一种新的码率控制算法,基于人眼显著性区域的检测,以优化视频编码效率。该算法利用像素级的恰可感知失真(JND)模型,精准预测人眼对图像失真感知的阈值。通过分析相邻帧间的像素值变化,识别出图像中的显著性区域,并基于JND模型进一步细分这些区域。通过构建一个多级码率分配框架,以指导不同区域的码率分配。实验结果显示,在低码率环境下,该算法能够显著提升显著性区域的视觉质量,同时保持码率控制的精确性。相比现有的HM16.9算法,在码率精确性基本不变的情况下,该文算法的BD-Rate下降0.028 9%,BD-PSNR提升0.001 6%,多个测试序列的结构相似性△SSIM平均提升0.000 69。This paper proposes a new bitrate control algorithm based on human visual saliency detection to optimize video coding efficiency.This algorithm utilizes a pixel-level Just Noticeable Distortion(JND)model to accurately predict the threshold for human eye perception of image distortion.By analyzing the pixel value changes between adjacent frames,salient regions in the image are identified,and these regions are further subdivided based on the JND model.By constructing a multi-level rate allocation framework to guide rate allocation in different regions.The experimental results show that in low-bitrate environments,this algorithm can significantly improve the visual quality of saliency while maintaining the accuracy of bit rate control.Compared with the existing HM16.9 algorithm,while the bit rate accuracy remains basically unchanged,the BD-Rate of the algorithm in this paper decreases by 0.0289%,the BD-PSNR increases by 0.0016%,and the average structural similarity?驻SSIM of multiple test sequences increases by 0.00069.

关 键 词：HEVC 显著性区域 码率控制 恰可感知失真模型 原始图像 

分 类 号：TN919.8[电子电信—通信与信息系统]