基于多节点拓扑重叠测度高阶MRF模型的图像分割

Image Segmentation Based on Higher-order MRF Model With Multi-node Topological Overlap Measure

针对低阶马尔科夫随机场(Markov random field, MRF)模型难以有效表达自然图像中复杂的先验知识而造成误分割问题,提出一种基于多节点拓扑重叠测度高阶MRF模型(Higher-order MRF model with multi-node topological overlap measure, MTOM-HMRF)的图像分割方法.首先,为描述图像局部区域内多像素蕴含的复杂空间拓扑结构信息,利用多节点拓扑重叠测度建立图像局部区域的高阶先验模型;其次,利用较大的局部区域包含更多的标签节点信息能力,基于Pairwise MRF模型建立基于局部区域的部分二阶Potts先验模型,提高分割模型的抗噪能力;再次,为有效描述观察图像场与其标签场的似然特征分布,研究利用局部区域内邻接像素的Hamming距离引入图像局部空间相关性,建立局部空间一致性约束的高斯混合分布;最后,基于MRF框架建立用于图像分割的多节点拓扑重叠测度高阶MRF模型,采用Gibbs采样算法对提出模型进行优化.实验结果表明,提出模型不仅能有效抵抗图像强噪声和复杂的纹理突变干扰,鲁棒性更好,而且具有更准确的图像分割结果.

Aim at the problem that lower-order Markov random field(MRF) model is inefficient to capture the rich prior knowledge of nature images which may bring out error image segmentation results, a new image segmentation method is proposed based on higher-order MRF model with multi-node topological overlap measure(MTOM-HMRF). Firstly, to capture complex spatial topological structure information embedded in the local region of images,the proposed method utilizes the topological overlap measure among multi-image-pixels to build higher-order prior model for the local region of images. Secondly, according that larger local region contains more information in label nodes, a partial 2-order Potts model is built based on pairwise MRF model, which increases the anti-noise capability of the proposed model. Thirdly, to efficiently describe the likelihood distribution between observed image field and its label field, a local spatial consistency constraints Gaussian mixture distribution is constructed based on the Hamming distribution between neighbor pixels which incorporated image local spatial correlation. Finally, a topological overlap measure higher-order MRF model is proposed for image segmentation based on the MRF framework,and Gibbs sampling algorithm is used to optimize the proposed model. Experimental results on artificial synthesis images and nature images show that the proposed model is not only efficient to overcome the impact of strong noise and complex texture abrupt on image segmentation results, thus possesses more robustness, but also can provide more accurate edge segmentation results.