基于各向异性Markov随机场的三维显微图像复原  

3-D microscopical image restoration based on anisotropic Markov random field in wavelet domain

作　　者：何小海[1] 张菊[1] 陶青川[1] 吕成淮[1] 王宇[1] 

机构地区：[1]四川大学电子信息学院图像信息研究所,四川成都610064

出　　处：《光学精密工程》2008年第4期733-739,共7页Optics and Precision Engineering

基　　金：国家自然科学基金资助项目(No.60372079)

摘　　要：提出了基于小波域各向异性马尔可夫随机场模型的三维显微图像复原算法。根据小波变换后各个子带内小波系数的方向性,以各向异性马尔可夫随机场模型作为小波系数的先验概率模型进行正则化处理,正则化比例系数采用自适应调整方法;考虑到噪声的影响,算法在每一步迭代求解过程中对估计出的图像进行去噪处理。实验结果表明,该算法能够有效地保留图像边缘等细节信息,去除层间干扰并抑制噪声。当算法收敛到全局最优时,正则化比例系数也达到了最优选择。与最小二乘共轭梯度法相比,信噪比和峰值信噪比均提高了1 dB以上。To preserve the edges of restored images, an algorithm based on anisotropic Markov Random Field （MRF） model for 3-D microscopical image restoration is proposed. Since the wavelet coefficients in different subbands have different orientational characteristics, a prior probability constraint based on anisotropic MRF is introduced as a prior penalized term in the regularization process. The regularization coefficients are adaptively updated according to the restored results during each iteration. By taking the noise effects into account, the estimated images are denoised in the iteration process. The experimental results show that the regularization parameters decrease step by step when the estimated images become close to the real ones. The new algorithm can effectively preserve the detail information, and satisfactorily reduce both the interruption between layers of 3-D slices and the noise effects. The ISNR and PSNR both increase more than i dB compared with those of the CGLS algorithm.

关 键 词：图像复原 边缘保持 各向异性马尔可夫随机场 小波变换 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]