基于改进Random Walk算法的磁共振图像脑组织分割

MRI Brain Tissue Segmentation Based on Improved Random Walk Algorithm

在Random Walk算法中,边的权重对算法分割性能有着重要的影响,针对原算法在计算权重时仅使用相邻像素灰度值变化信息的局限性,通过引入图像局部熵,使得权重函数同时反映相邻像素灰度变化信息和图像局部灰度离散度信息,增强算法对目标内容和边界的识别能力。同时通过Fisher评价函数构造最佳分类阈值的选取法则,增强了算法的判别能力。实验表明改进后的算法对目标内容和边界的识别能力有较大提高,且对噪声具有更好的鲁棒性。

The edge weight is the most important factor in Random Walk algorithm.To improve the mapping ability of the weighting function can achieve better performance.The image local entropy is introduced into Random Walk algorithm to construct a new weighting function reflecting the changing information of adjacent-pixel′s gray value and the discrete information of local image.It improves the identification ability on homogeneous pixels and edge.The Fisher discriminate function is used to calculate the optimal classification threshold.Experimental results show that the improved algorithm performs better for identifying homogeneous content and boundary of target,and it has a better robustness to noise than the original algorithm.