摘要： 核磁共振等医学图象，由于受成象方法的限制，图象中病变部位常湮没在噪声信号中，引起医生的漏诊甚至误诊。如果经过适当降噪预处理，可以使病变部位特征更加明显，从而使医生的诊断更准确。但降噪算法常会因过度平滑引起图象细节特征的模糊，因此需要采用能保留细节的平滑处理算法。该文采用改进的基于梯度的偏微分算法，克服了退化扩散方程对图象细节的模糊效应，兼顾了降噪和保细节。对比几种常用的保边界平滑算法，实验结果证明，该文的改进算法对于核磁共振脑瘤图象的预处理，优于其他几种算法。

关键词: 图象平滑, 保细节平滑, 偏微分方程, 退化扩散方程, 核磁共振图象

Abstract: It is necessary for MRI(Magnetic Resonance Image) to be preprocessed before being used for analysis due to the limitation of imaging methodology, since the abnormal symptoms submerged in noise may lead to missed diagnosis or even misdiagnosis. However, smoothing often brings about blurred image. Although therere several kinds of edge preserving smoothing algorithms, the results are not satisfying. In this paper, we employ an improved minutia preserving smoothing algorithm based on solving a nonlinear diffusion equation. This method, to some extent, is effective for both smoothing and minutia preserving, mainly because it prevents the blurring effect of degenerate diffusion equation for these points with big gray level variations. Compared with other methods, this algorithm has better performance for preprocessing of noisy MRI images.

中图分类号: 

• TP391.4