显著图联合改进的JND模型

Improved Just Noticeable Model with Saliency Map

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.

Based on existing saliency detection algorithm, a model of joint estimation just noticeable distortion（JND） was proposed.This model not only makes full use of the advantage of the JND model, based on pixel domain and transform domain, but also employs the character of human visual system.The saliency detection algorithm used for saliency map, is employed to assign different weights to JND model, based on different brightness and color.The JND model based on pixel domain is mainly used on plain area, and the JND model, with masking effect and contrast sensitivity function, based on transform domain is used on texture area.The proposed model is introduced into the HEVC sample platform HM16.4.Experiments show that the proposed JND model, compared to HEVC standard, have the similar subjective quality.