磁共振图像的降噪处理一直是医学图像处理中重要的研究领域。图像中存在噪声会降低图像质量从而影响临床诊断。现有K-SVD 算法虽然能达到良好的去噪效果，但却在字典训练中消耗大量时间。本文针对时间消耗问题，提出利用改进的KSVD算法进行医学图像去噪。首先根据已知的字典原子的可稀疏性，提出一种高效、灵活的稀疏字典结构，该字典能够提供高效的前向和伴随算子，并具有紧凑的表示形式，同时可以有效地训练图像信号；然后在现有K-SVD 算法的基本框架下，结合字典的稀疏表示特点使用改进K-SVD 算法训练稀疏字典，改进的K-SVD 算法能够对更大的字典进行训练，特别是对高维数据的处理更具有优势。实验结果表明，该算法相对基于离散余弦变换字典的磁共振图像去噪以及基于传统K-SVD 算法的磁共振图像去噪，不仅能够更加有效地滤除图像中的高斯白噪声，更好地保留原图像的细节信息，而且有效降低了字典训练所消耗的时间；在相同的噪声标准差下，改进K-SVD 算法的峰值信噪比提高了约1~3 dB。
Magnetic resonance image is an important research field in medical image processing. Because it can degrade the image quality, the signal noises have a negative impact on clinical diagnosis. The K-SVD algorithm can obtain better de-noising results, but the time-consuming problem of the dictionary training still exists. A medical image denoising algorithm based on improved K-SVD is studied to solve this problem. First of all, an efficient and flexible dictionary structure is proposed based on a sparsity model of the dictionary atoms over a know dictionary. The sparse dictionary provides efficient forward and adjoint operators, has a compact representation, and can be effectively trained from given example data. Then the basic framework of the existing K-SVD algorithm, combined with the dictionary sparse representation, can improve K-SVD training algorithm, and the improved K-SVD algorithm can be trained for greater dictionary, especially for high-dimensional data. Therefore, it can be used to remove the noise of magnetic resonance images. The experimental results show that the algorithm, in comparison with the discrete cosine transform dictionary and conventional K-SVD algorithm, can effectively filter Gaussian white noise of the image to retain image details, and reduce the time of dictionary training. It is found that the peak signal-to-noise ratio is increased by about 1~3db with the proposed method.
[1] Oster J, Pietquin O, Kraemer M, et al. Nonlinear bayesian filtering for denoising of electrocardiograms acquired in a magnetic resonance environment[J]. IEEE Transaction on Biomedical Engineering, 2010, 57 (7): 1628-1638.
[2] Coupé P, Yger P, Prima S, et al. An optimized blockwise nonlocal means denoising filter for 3-D magnetic resonance images[J]. IEEE Transactions on Image Processing, 2008, 27(4): 425-441.
[3] Wang Y, Che X Q, Ma S L. Nonlinear filtering based on 3D wavelet transform for MRI denoising[J]. EURASIP Journal on Advances in Sinal Processing, 2012, 40: 1-14.
[4] Ong F, Uecker M, Tariq U, et al. Improved visualization and quantification of 4D flow MRI data using divergence- freewavelet denoising[C]//IEEE 10th International Symposium on Biomedical Imaging. San Francisco, CA, USA, April 7-11, 2013.
[5] Donoho D. Compressed sening[J]. IEEE Transactions on Information Theory, 2006, 52(4):1289-1306.
[6] Donoho D, Reeves G. The sensitivity of compressed sensing performance to relaxation of sparsity[C]//IEEE International Symposium on Information Theory Proceedings. California, MA, USA, July 1-6, 2012.
[7] 邢博, 王俊. 基于稀疏分解的医学CT图像去噪[J]. 生物医学工程学杂 志, 2012, 29(3): 456-459. Xing Bo, Wang Jun. Denoising of medical CT image based on sparse decomposition[J]. Journal of Biomedical Engineering, 2012, 29(3): 456-459.
[8] 郭德全, 杨红雨, 刘东权, 等. 基于稀疏性的图像去噪综述[J]. 计算机 应用研究, 2012, 29(2): 406-412. Guo Dequan, Yang Hongyu, Liu Dongquan, et al. Overview on sparse image denoising[J]. Application Research of Computer, 2012, 29(2): 406-412.
[9] Mallat S, Zhang Z. Matching pursuits with time-frequency dictionaries[J]. IEEE Transaction on Image Processing, 1993, 41(12): 3397-3415.
[10] Tropp J. Greed is good: Algorithm results for sparse approximation[J]. IEEE Transaction Information Theory, 2004, 50(10): 2231-2242.
[11] Tropp T A, Gilbert A, Muthukrishnan S, et al. Improved sparse approximation over quasiincoherent dictionaries[C]//2003 IEEE International Conference on Image Processing. Barcelona, Spain, Sept. 14-17. 2003.
[12] Candès E J, Donoho D L. New tight frames of curvelets and optimal representations of objects with piecewise C2 singularities[J]. Communications on Pure and Applied Mathematics, 2004, 57(2): 219- 266.
[13] Aharon M, Elad M, Bruckstein A. K-SVD: An algorithm for designing overcomplete dictionaries for sparse representation[J]. IEEE Transactions on Signal Processing, 2006, 54(11): 4311-4322.
[14] Elad M, Aharon M. Image denoising via sparse and redundant representations over learned dictionaries[J]. IEEE Transactions on Image Processing, 2006, 15(12): 3736-3745.
[15] Hu Jinrong, Pu Yifei, Wu Xi, et al. Improved DCT- based nonlocalmeans filter for MR images denoising[J]. Computational and Mathematical Methods in Medicine, 2012, 12(1): 1-13.
[16] Rubinstein R, Peleg T, Elad M. Analysis K- SVD: A dictionarylearning algorithm for the analysis sparse model[J]. IEEE Transactions on Signal Processing, 2013, 61(3): 661-677.
