Many neuroanatomy studies rely on brain tissue segmentations of magnetic resonance images. In order to segment these images, many active contour methods have been presented. But the traditional active contour method only uses the information of the edge, when it segment magnetic resonance images with strong noise or weak edges, which is popular in medical images, so it is difficult to get the true edge. In this paper the Gaussian mixture model is used to make a new sanction. With this sanction the model can reduce the effect of the noise and prevent the curve over the edge. The expectation-maximization (EM) method is the popular method to solve the Gaussian mixture model, but it is a local optimizer method and is sensitive to the initial value. The global optimization characteristic of the particle swarm optimizer method, which is based on a metaphor of social interaction, is used to solve this problem. The classical particle swarm optimizer method is sensitive to the initial location. In order to overcome this problem, Powell method and new corrupt method are used to adapt the particle swarm optimizer method and with the new adapted particle swarm optimizer method the Gaussian mixture model can get global best results. Experiments on the segmentation of brain magnetic resonance images show that the proposed model can gain better results in image segmentation.