计算机工程

• 图形图像处理 • 上一篇    下一篇

基于核稀疏表示的图像去噪算法

韩金菊,邹国良   

  1. (上海海洋大学信息学院,上海 201306)
  • 收稿日期:2015-01-22 出版日期:2016-03-15 发布日期:2016-03-15
  • 作者简介:韩金菊(1988-),女,硕士研究生,主研方向为图形图像处理、信息安全;邹国良,教授、博士。
  • 基金项目:
    河口海岸学国家重点实验室开放课题基金资助项目(SKLEC201207)。

Image Denoising Algorithm Based on Kernel Sparse Representation

HAN Jinju,ZOU Guoliang   

  1. (College of Information Technology,Shanghai Ocean University,Shanghai 201306,China)
  • Received:2015-01-22 Online:2016-03-15 Published:2016-03-15

摘要: 传统去噪算法去除噪声后仍有噪声残留,且噪声较大时的图像去噪效果不明显。针对该问题,提出一种新的图像去噪算法。将输入的噪声图像分成相互重叠的图像块,随机抽取适量的图像块学习得到自适应的冗余字典,给出核正则化正交匹配追踪技术,利用该技术得到稀疏表示系数,并使用稀疏表示系数恢复原图像。实验结果表明,与K-奇异值分解算法相比,该算法的峰值信噪比较高,且能较好地保持图像的细节和纹理信息。

关键词: 字典学习, 冗余字典, 核稀疏表示, 图像去噪, 正交匹配追踪

Abstract: The traditional denoising algorithm has residual noise after removing noise,and image denoising effect is not obvious for the large noise.Aiming at this problem,a new image denoising algorithm is proposed in this paper.In this algorithm,the input image with noise can be split into overlapped image patches.Through randomly selecting moderate image block to learn,an adaptive redundant dictionary can be got.Then the sparse representation coefficients can be obtained from this redundant dictionary with nuclear regularized orthogonal matching pursuit technology.Then the image can be restored by these coefficients.Experimental results show that compared with K-Singular Value Decomposition(K-SVD) algorithm,the Peak Signal to Noise(PSNR) of the proposed algorithm is better,the image detail and texture information can be well preserved.

Key words: dictionary learning, redundant dictionary, kernel sparse representation, image denoising, Orthogonal Matching Pursuit(OMP)

中图分类号: