局部全变分卡通纹理分解医学影像滤波

Medical image filtering based on local total variation cartoon-texture decomposition

针对直接采用全局滤波方法会模糊图像的边缘,丢失重要纹理信息的弊端,提出一种基于快速变分卡通纹理分解的非局部正则化滤波方法。首先将非线性局部全变分作为图像分解的指示函数,分离图像的卡通结构分量和纹理震荡分量;其次根据震荡分量极强的重复性及结构方向性,结合具有较好方向选择性以及块相似匹配性的非局部梯度正则项,利用分裂Bregman方法求解震荡分量的非局部全变分极小化问题;最后将分解得到的卡通分量与去除噪声的震荡分量进行加权合成,得到复原图像。实验结果表明该算法对有精细结构和纹理的医学图像具有更佳的恢复效果,可以有效地应用于临床诊断以及后续分割。

In view of the disadvantages of the direct use of global filtering method,blurring image edges and losing important texture information,a nonlocal regularization filtering method based on fast variation cartoon-texture decomposition is proposed.Firstly,the nonlinear local total variation is used as an indicator function of image decomposition for separating the cartoon structural component and texture vibration component of the image.Secondly,according to the outstanding repeatability and structural orientation of the vibration component,a nonlocal gradient regular term with preferable directional selectivity and block similarity matching is combined with split Bregman method to solve the nonlocal total variation minimization problem of the vibration component.Finally,the combination of the decomposed cartoon component and the vibration component after noise removal is used for obtaining the restored image.The experimental results show that the proposed algorithm has excellent recovery effects for medical images with fine structure and texture.The proposed method can be effectively applied to clinical diagnosis and subsequent segmentation.