西安电子科技大学学报 ›› 2016, Vol. 43 ›› Issue (2): 120-125.doi: 10.3969/j.issn.1001-2400.2016.02.021

• 研究论文 • 上一篇    下一篇

非凸混合总变分图像盲复原

刘巧红1;李斌1;林敏2   

  1. (1. 上海大学 机电工程与自动化学院,上海  200073;
    2. 上海医疗器械高等专科学校 医学电子与信息工程系,上海  200093)
  • 收稿日期:2014-12-23 出版日期:2016-04-20 发布日期:2016-05-27
  • 通讯作者: 刘巧红
  • 作者简介:刘巧红(1979-),女,讲师,上海大学博士研究生,E-mail:hqllqh@163.com.
  • 基金资助:

    上海市教育委员会科研创新资金资助项目(14YZ169)

Non-convex hybrid total variation method for image blind restoration

LIU Qiaohong1;LI Bin1;LIN Min2   

  1. (1. School of Mechatronic Engineering and Automation, Shanghai Univ., Shanghai  200073, China;
    2. Department of Medical Electronics and Information Engineering, Shanghai Medical Instrumentation College, Shanghai  200093, China)
  • Received:2014-12-23 Online:2016-04-20 Published:2016-05-27
  • Contact: LIU Qiaohong

摘要:

为实现模糊噪声图像的盲复原,提出了一种混合非凸总变分和高阶总变分的多正则化约束的图像盲复原方法.首先,根据自然图像边缘的稀疏特性,运用了非凸总变分对复原图像进行正则化约束;然后,结合高阶总变分正则化克服阶梯效应的优势,建立了非凸混合总变分极小化模型;最后,利用增广拉格朗日方法和新的广义p收缩算子对提出的模型进行最优化求解.实验结果表明,提出的方法能够有效保护图像边缘细节,同时消除了图像平滑区域的阶梯效应,获得高质量的复原图像.

关键词: 图像复原, 非凸, 高阶, 总变分, 增广拉格朗日方法, p收缩算子, 优化

Abstract:

A multi-regularization constraint method for imageblind restoration is proposed to recover the blurry-noisy images.First, the non-convex total variation is adoptedas the regularization constraint by taking the sparse edges in the natural image into consideration. Next, the high-order total variation is used to overcome the staircase effects in the smooth regions of the image. Then a non-convex minimization model is proposed. Finally, the augmented Lagrangian method and a new generalized p shrinkage operator are applied to solve the model. The results of numerical experiments show that the proposed method can preserve the image edges while removing the staircase effects effectively. The high quality restored image can be obtained.

Key words: image restoration, non-convex, high-order, total variation, augmented Lagrangian method, p shrinkage operator;optimization

Baidu
map