计算机应用 ›› 2015, Vol. 35 ›› Issue (2): 506-509.DOI: 10.11772/j.issn.1001-9081.2015.02.0506

• 虚拟现实与数字媒体 • 上一篇    下一篇

基于自适应约束正则HL-MRF先验模型的MAP超分辨率重建

秦龙龙1, 钱渊1, 张晓燕1, 侯雪2, 周芹1   

  1. 1. 空军工程大学 信息与导航学院, 西安 710077;
    2. 空军哈尔滨飞行学院, 哈尔滨 150001
  • 收稿日期:2014-09-13 修回日期:2014-11-17 出版日期:2015-02-10 发布日期:2015-02-12
  • 通讯作者: 秦龙龙
  • 作者简介:秦龙龙(1988-),男,陕西咸阳人,硕士研究生,主要研究方向:视频图像清晰化、图像超分辨率重建; 钱渊(1972-),男,上海人,副教授,硕士,主要研究方向:图像处理、多媒体通信; 张晓燕(1970-),女,陕西西安人,副教授,博士,主要研究方向:多媒体信息融合;侯雪(1980-),女,辽宁沈阳人,助理工程师,主要研究方向:图像处理; 周芹(1990-),女,江苏东海人,硕士研究生,主要研究方向:视频图像清晰化、图像超分辨率重建。
  • 基金资助:

    陕西省自然科学基金资助项目(2013JM8025)。

MAP super-resolution reconstruction based on adaptive constraint regularization HL-MRF prior model

QIN Longlong1, QIAN Yuan1, ZHANG Xiaoyan1, HOU Xue2, ZHOU Qin1   

  1. 1. College of Information and Navigation, Air Force Engineering University, Xi'an Shaanxi 710077, China;
    2. Air Force Harbin Flight Academy, Harbin Heilongjiang 150001, China
  • Received:2014-09-13 Revised:2014-11-17 Online:2015-02-10 Published:2015-02-12

摘要:

针对Huber-MRF先验模型对图像高频噪声抑制能力较差,而Gauss-MRF先验模型对图像高频过度惩罚的问题,提出了一种改进的自适应约束正则HL-MRF先验模型。该模型将Huber边缘惩罚低频函数与Lorentzian边缘惩罚高频函数相结合,对低频进行线性约束的同时对高频实现平滑惩罚;并采用自适应约束方法确定正则化参数,从而得到最优的参数解。与基于Gauss-MRF先验模型和Huber-MRF先验模型的超分辨率算法相比,HL-MRF先验模型获得的超分辨率重建图像在峰值信噪比(PSNR)和细节方面都有一定程度的提高,在抑制高频噪声、避免图像细节被过度平滑方面具有一定的优势。

关键词: 超分辨率重建, 马尔可夫随机场先验模型, 自适应正则化, 边缘惩罚函数

Abstract:

Aiming at the poor suppression ability for the high-frequency noise in Huber-MRF prior model and the excessive punishment for the high frequency information of image in Gauss-MRF prior model, an adaptive regularization HL-MRF model was proposed. The method combined low frequency function of Huber edge punishment with high frequency function of Lorentzian edge punishment to realize a linear constraint for low frequency and a less punishment for high frequency. The model gained its optimal solution of parameters by using adaptive constraint method to determine regularization parameter. Compared with super-resolution reconstruction methods based on Gauss-MRF prior model and Huber-MRF prior model, the method based on HL-MRF prior model obtains higer Peak Signal-to-Noise Ratio (PSNR) and better performace in details, therefore it has ceratin advantage to suppress the high frequency noise and avoid excessively smoothing image details.

Key words: super-resolution reconstruction, Markov Random Field (MRF) prior model, adaptive regularization, edge penalty function

中图分类号: