Abstract: The lattice Boltzmann (LB) model has advantages of simple programming and faster operation, but for images with low contrast and noise, segmentation may fail. This paper proposes a novel LB model using local statistical region information. As the method can enhance contrast of the object and background, and reduce noise, it provides improved delineation accuracy. To verify effectiveness of the model, comparison experiments among the existing LB model, level set models and the proposed model are made, using real magnetic resonance (MR) images. Dice coefficient and Hausdorf distance are used as the measurement index. The results show that the proposed model produces segmentation results with precision 10 times better than the existing LB method. In addition, the computing speed is 3 times faster than level set models.

Key words: image segmentation, brain magnetic resonance image, lattice Boltzmann model, level set method