一种基于视觉特性的部分DCT系数重构算法
A Reconstruction algorithm from partial DCT coefficients based on human perception
郑明魁 1苏凯雄 1杨秀芝1
作者信息
- 1. 福州大学物理与信息工程学院,福建福州350108
- 折叠
摘要
基于全变差模型的图像重构算法对于处理图像边界问题有较好的优势,本文结合小波变换域稀疏和图像局部光滑的特性,使用基于全变差的混合模型对图像进行重构,并在重构时引入人眼视觉特性,根据人眼对不同频率的系数具有不同的敏感度,提出了一种不等概率的DCT部分系数重构算法.通过对不同频率的系数进行不等概率的采样,低频系数以较高概率采样,仿真结果表明,与现有的一些重构算法相比,在相同的测量点数下,重构图像质量得到较大提高.
Abstract
Total Variation (TV) regularization is widely exploited in image reconstruction problem due to the advantage of preserving image's edges.Numerous practical methods are proposed recently.In this paper,the hybrid regularization principle based on Total Variation is introduced to reconstruct the image.Human eyes have different sensitivity for different frequency coefficients,typically,low frequency components are more sensitive than high frequency components.According to the perceptual properties of human eyes,we propose an improved reconstruction algorithm based on unequal probability sampling for the partial DCT coefficients in this paper.We adjust low frequency coefficients higher probability than high frequency components in the sampling,because low frequency components are more important than high frequency components to visual quality,so that the reconstructed image quality would be enhanced.Compared with the TVAL and IADPM algorithm,experimental results show that our proposed algorithm can enhance the quality of the recovered image significantly at the same sampling rate.
关键词
全变差/混合模型/视觉特性/不等概率采样Key words
Total Variation (TV)/hybrid regularization/human perception/unequal probability sampling引用本文复制引用
基金项目
国家自然科学基金(61170147)
福建省高校产学合作重大项目(2012H61010016)
福建省科技厅产学研重点项目(2012H6012)
出版年
2013
NETL
NSTL
维普
万方数据