电信科学 ›› 2023, Vol. 39 ›› Issue (2): 59-70.doi: 10.11959/j.issn.1000-0801.2023014

• 研究与开发 • 上一篇    下一篇

基于熵掩蔽的DCT域恰可察觉失真模型

骆琼华, 王鸿奎, 殷海兵, 邢亚芬   

  1. 杭州电子科技大学通信工程学院,浙江 杭州 310018
  • 修回日期:2023-01-09 出版日期:2023-02-20 发布日期:2023-02-01
  • 作者简介:骆琼华(1998- ),女,杭州电子科技大学通信工程学院硕士生,主要研究方向为视频感知编码
    王鸿奎(1990- ),男,博士,杭州电子科技大学通信工程学院讲师,主要研究方向为视觉感知理论、视频感知编码以及视频智能编码
    殷海兵(1974- ),男,博士,杭州电子科技大学通信工程学院教授,主要研究方向为数字视频编解码、图像和视频处理以及VLSI结构设计
    邢亚芬(1997- ),女,杭州电子科技大学通信工程学院硕士生,主要研究方向为视频感知编码
  • 基金资助:
    国家自然科学基金资助项目(62202134);国家自然科学基金资助项目(61972123);国家自然科学基金资助项目(61931008);国家自然科学基金资助项目(62031009);浙江省“尖兵”“领雁”研发攻关计划项目(2023C01149);浙江省“尖兵”“领雁”研发攻关计划项目(2022C01068)

Just noticeable distortion model based on entropy masking in DCT domain

Qionghua LUO, Hongkui WANG, Haibing YIN, Yafen XING   

  1. College of Communication Engineering, Hangzhou Dianzi University, Hangzhou 310018, China
  • Revised:2023-01-09 Online:2023-02-20 Published:2023-02-01
  • Supported by:
    The National Natural Science Foundation of China(62202134);The National Natural Science Foundation of China(61972123);The National Natural Science Foundation of China(61931008);The National Natural Science Foundation of China(62031009);Zhejiang Provincial “Pioneer” and “Leading Goose” Research and Development Project(2023C01149);Zhejiang Provincial “Pioneer” and “Leading Goose” Research and Development Project(2022C01068)

摘要:

为提高离散余弦变换(discrete cosine transform,DCT)域恰可察觉失真(just noticeable distortion, JND)模型阈值精度并避免跨域操作,将熵掩蔽效应引入DCT域JND模型。首先,从自由能理论和贝叶斯推理出发,设计基于DCT域纹理能量相似性的自回归模型模拟视觉感知过程中的自发预测行为;其次,探索视觉感知与预测残差的映射关系得到块级无序度,并将熵掩蔽效应建模为关于无序度的JND阈值调节因子;最后,结合空间对比敏感度函数、亮度自适应掩蔽以及对比度掩蔽,提出基于熵掩蔽的 DCT 域 JND 模型。与现有DCT域JND模型相比,所提模型所有运算均在DCT域执行,更高效简洁。主观、客观实验结果表明,所提模型在感知质量相同或更好的情况下,噪声污染图的平均峰值信噪比(peak signal-to-noise ratio,PSNR)值比其他4个JND对比模型低2.04 dB,更符合人眼视觉系统的感知特性。

关键词: 恰可察觉失真, 人眼视觉系统, 熵掩蔽效应, 自由能理论, 贝叶斯推理

Abstract:

In order to improve the threshold accuracy of JND (just noticeable distortion) model in DCT (discrete cosine transform) domain and avoid cross-domain operation, entropy masking effect was introduced into DCT-based JND model.Firstly, starting from the free-energy theory and the Bayesian inference, an autoregressive model based on texture-energy similarity in DCT domain was designed to simulate the spontaneous prediction behavior of visual perception.Secondly, the mapping relationship between visual perception and prediction residuals were explored to obtain the disorder intensity in block level.Thirdly, the entropy masking effect was modeled as a JND threshold modulation factor of disorder intensity.Finally, the JND model in DCT domain for the entropy masking was proposed by fusing the contrast sensitivity function, the luminance adaptive masking, the contrast masking.Compared with the existing JND model in DCT domain, the proposed model performed all operations in DCT domain, which was more efficient and concise.The subjective and objective experimental results indicate that the proposed JND model shows greater tolerance to distortion with better perceptual quality.

Key words: JND, human visual system, entropy masking effect, free-energy theory, Bayesian inference

中图分类号: 

No Suggested Reading articles found!