计算机应用 ›› 2015, Vol. 35 ›› Issue (4): 1075-1078.DOI: 10.11772/j.issn.1001-9081.2015.04.1075

• 虚拟现实与数字媒体 • 上一篇    下一篇

基于带间预测的非负支撑域受限递归逆滤波盲复原算法

黄德天, 郑力新, 柳培忠, 顾培婷   

  1. 华侨大学 工学院, 福建 泉州 362021
  • 收稿日期:2014-11-03 修回日期:2014-12-19 出版日期:2015-04-10 发布日期:2015-04-08
  • 通讯作者: 黄德天
  • 作者简介:黄德天(1985-),男,福建龙岩人,讲师,博士,主要研究方向:图像复原、视频调焦、嵌入式系统; 郑力新(1976-),男,福建泉州人,教授,博士,主要研究方向:自动化技术、人工智能、机器视觉; 柳培忠(1976-),男,福建泉州人,讲师,博士,主要研究方向:仿生图像处理、模式识别; 顾培婷(1991-),女,福建泉州人,硕士研究生,主要研究方向:图像盲复原。
  • 基金资助:

    国家自然科学基金资助项目(61203242);华侨大学科研基金资助项目(13BS416);物联网云计算平台建设项目(2013H2002);泉州市科技计划项目(2014Z113)。

Improved non-negativity and support constraint recursive inverse filtering algorithm for blind restoration based on interband prediction

HUANG Detian, ZHENG Lixin, LIU Peizhong, GU Peiting   

  1. College of Engineering, Huaqiao University, Quanzhou Fujian 362021, China
  • Received:2014-11-03 Revised:2014-12-19 Online:2015-04-10 Published:2015-04-08

摘要:

针对非负支撑域受限递归逆滤波(NAS-RIF)算法对噪声敏感和耗时长等缺点,提出了一种改进的NAS-RIF盲复原算法。首先,为了改进原始NAS-RIF算法的抗噪性能和复原效果,引入了一种新的NAS-RIF算法代价函数;其次,为了提高算法的运算效率,结合Haar小波变换,仅对低频子频带的图像进行NAS-RIF算法复原,而高频子频带的信息,则通过带间预测分别从低频子频带的复原图像中预测得到;最后,为了保证高频信息的准确性,提出了一种基于最小均方误差(MMSE)的带间预测。分别对模拟退化图像和真实图像进行了仿真实验,采用该算法得到的信噪比增益分别为5.2216 dB和8.1039 dB。实验结果表明:该算法在保持图像边缘细节的前提下,能够较好地抑制噪声;此外,该算法的运算效率也得到了较大的提高。

关键词: 图像盲复原, 非负支撑域受限递归逆滤波算法, Haar小波变换, 带间预测

Abstract:

To overcome the shortcoming that the Non-negativity And Support constraint Recursive Inverse Filtering (NAS-RIF) algorithm is noise-sensitive and time-consuming, an improved NAS-RIF algorithm for blind restoration was proposed. Firstly, a new cost function of the NAS-RIF algorithm was introduced, and then the noise resistance ability and the restoration effect were both improved. Secondly, in order to enhance computational efficiency of the algorithm, after decomposed by Haar wavelet transform, only degraded image in low frequency sub-bands was restored with the NAS-RIF algorithm, while information in high frequency sub-bands was predicted from the restored image of low frequency sub-bands by interband prediction. Finally, an interband prediction based on Minimum Mean Square Error (MMSE) was presented to guarantee the accuracy of the predicted information in high frequency sub-bands. The experiments on synthetic degraded images and real images were performed, and the Signal-to-Noise Ratio (SNR) gain by proposed algorithm were 5.2216 dB and 8.1039 dB respectively. The experimental results demonstrate that the proposed algorithm not only preserves image edges, but also has good performance in noise suppression. In addition, the computational efficiency of the proposed algorithm is greatly enhanced.

Key words: blind image restoration, Non-negativity And Support constraint Recursive Inverse Filtering (NAS-RIF) algorithm, Haar wavelet transform, interband prediction

中图分类号: