Abstract: MRI（Magnetic Resonance Imaging） is easily affected by noise, and it has poor contrast along boundaries. MRI of brain tissue segmentation under the strong noise has always been a difficult problem, and it has attracted much attention. This paper puts forward a kind of algorithm using adaptive regularization parameters combined with spatial relation, which replaces the Euclidean distance by the Kernel distance for calculation, and segments the MR image under the strong noise, the robustness of segmentation is greatly improved. The main advantage is to define adaptive parameters for each point, and puts the parameters into two expressions of the objective function. And it not only reduces the number of parameters, but also enhances the segment result. Finally, combined with spatial relation, the segmentation is more accurate. The experiments show the proposed method improves the segmentation accuracy, detail retention and noise processing in brain.

Key words: Magnetic Resonance Imaging（MRI）, adaptive parameter, kernel distance, spatial relation

摘要： 核磁共振图像（Magnetic Resonance Imaging）容易受到噪声的干扰，并且在图像边缘部分呈弱对比度。强噪声下核磁共振图像的脑组织分割一直是个难题，引起很多学者的关注。提出了一种使用自适应正则化参数并结合空间关系的算法，同时将核距离替换传统的欧式距离进行计算，对强噪声下的核磁共振图像进行分割，大大提高了分割的鲁棒性。算法的主要优点是为图像每个点定义自适应参数，并且将这个参数同时应用到目标函数的两项表达式当中，既减少了参数数量，又增强了分割效果。最后，由于结合空间关系，使分割结果更加的精确。实验表明，该方法在脑组织的分割精度、细节保留以及噪声处理方面比其他方法有所提高。

关键词: 核磁共振图像, 自适应参数, 核距离, 空间关系