论文部分内容阅读
提出一种改进的基于显著性检测图联合估计恰可失真(JND)阈值的视觉感知模型,将人眼注意力机制引入JND模型,通过感知特点建模得到更为精确的JND模型.首先通过改进的显著性检测算法得到相应的显著图,在计算JND阈值的过程中,使用显著图来分配不同的权重给JND模型,并针对色度和亮度的不同给予不同的权重.基于空域的JND模型主要用在计算图像中的平坦区域;而基于DCT域的JND模型更加适合计算纹理区域的阈值,新的模型还同时考虑加入对比敏感度函数和各种掩蔽效应因子.将改进的JND模型融合到新的视频编码软件HM16.4中,实验结果表明,与HEVC标准的数据对比,视觉感知质量没有明显下降.
This paper proposes an improved visual perception model based on the saliency map and the JND threshold, introduces the human attention mechanism into the JND model, and through the perceptual feature modeling, obtains a more accurate JND model.At first, , The saliency map is used to get the saliency map. In the process of calculating JND threshold, saliency maps are used to assign different weights to the JND model, and different weights are given for different chroma and lightness. The JND model based on DCT domain is more suitable for calculating the threshold value of the texture region, and the new model also considers the contrast sensitivity function and various masking effect factors at the same time. The improved JND model is fused to the new Video coding software HM16.4, the experimental results show that, compared with the data of the HEVC standard, the visual perception of quality did not decline significantly.