传统的高斯混合模型(Gaussian mixture model, GMM)算法在图像分割中未考虑像素的空间信息,导致其对于噪声十分敏感。马尔科夫随机场(Markov random field, MRF)模型通过像素类别标记的Gibbs 分布先验概率引入了图像的空间信息,能较好地分割含有噪声的图像,然而MRF 模型的分割结果容易出现过平滑现象。为了解决上述缺陷,提出了一种新的基于图像片权重方法的马尔科夫随机场图像分割模型,对邻域内的不同图像片根据相似度赋予不同的权重,使其在克服噪声影响的同时能保持图像细节信息。同时,采用KL 距离引入先验概率与后验概率关于熵的惩罚项,并对该惩罚项进行平滑,得到最终的分割结果。实验结果表明,算法具有较强的自适应性,能够有效克服噪声对于分割结果的影响,并获得较高的分割精度。%Without considering the spatial information between pixels, the traditional Gaussian mixture model (GMM) algorithm is very sensitive to noise during image segmentation. Markov random field (MRF) models provide a powerful way to noisy images through Gibbs joint probability distribution which introduce the spatial information of images. However, they often lead to over-smoothing. To overcome these drawbacks, we propose a new brain MR image segmentation algorithm based on MRF with image patch by assigning each pixel in the neighborhood with a different weight according to the similarity between image patches. The proposed method can overcome the noise and keep the details of topology and corner regions. Meanwhile, by introducing the KL distance into the prior probability and posterior probability as an entropy penalty, the proposed algorithm could get better segmentation results through smoothing this penalty term. Experimental results show that our algorithm can overcome the impact of noise on the segmentation results adaptively and efficiently, and get accurate segmentation results.
展开▼
