Abstract: Segmenting brain from non-brain tissue，also know as skull stripping，has become an important image processing step in analyses involving image registration or cortical flattening.But with the effect of the bias fields，weak edges，strong noise，the level set method，only uses the information of the edges，can not get well results.Gaussian mixture model uses the global information of the image，so it can deal with the problems of the weak edges.But the traditional Gaussian mixture model only uses the information of the histogram and does not take the spatial information into account，so the model can not do well with noisy images.This paper constructs a new information field.Based on the new field information the Gaussian mixture model can reduce the effect of the bias field，noise，and prevent the noise covering the weak edges effectively.Gaussian mixture model can be solved by EM method，which can only get the local best results.In order to get the global results，particle swarm optimizer is introduced and improved to get result exactly and quickly.Experiments on the segmentation of brain magnetic resonance images show this model has better effect in image segmentation.

Key words: Gaussian mixture model, Expection-Maximization（EM） algorithm, particle swarm optimizer, image segmentation, nuclear Magnetic Resonance Imaging（MRI）

摘要： 将脑部组织从MR图像中提取出来已经成为脑部图像处理中的一个重要环节，它可以提高后继的脑组织定位、容积测量等处理的精确度。但由于脑MR图像往往具有偏移场、弱边界和强噪音，使得基于图像梯度信息的水平集模型很难得到真实解。高斯混合模型使用了图像全局信息，能较好地处理弱边界问题。但传统的高斯混合模型仅使用了灰度值分布信息，未对像素的位置进行考虑，这使得其在处理噪音图像时效果并不是很理想。利用图像多种信息构造新的信息场，使得由信息场构造的高斯混合模型更能降低偏移场、噪音等影响，同时防止曲线从弱边界泄漏。传统的高斯混合模型求解参数时，往往仅使用EM算法，易陷入局部最优。针对这个缺点，引入粒子群算法，并对其进行改进，使得改进的算法可以较快地得到精确解。对脑MR图像分割实验表明该模型可得到较好的分割效果。

关键词: 高斯混合模型, EM算法, 粒子群算法, 图像分割, 核磁共振成像