A region Markov random field model with integrated edge feature and image segmentation algorithm

被引：0

作者：

MOE Key Lab for Intelligent Networks and Network Security, Xi'an Jiaotong University, Xi'an 710049, China ^{[1
]}

不详 ^{[2
]}

机构：

[1] MOE Key Lab for Intelligent Networks and Network Security, Xi'an Jiaotong University

[2] School of Information and Control Engineering, Xi'an University of Architecture and Technology

来源：

Han, J. | 1600年 / Xi'an Jiaotong University卷 / 48期

关键词：

Edge feature; Gaussian mixture model; Image segmentation; Markov random field;

D O I：

10.7652/xjtuxb201402003

中图分类号：

学科分类号：

摘要：

A region Markov random field model with integrated edge feature (IEFRMRF) is proposed to solve the problem that existing region Markov random fields (MRF) model often leads to produce blur edge in image segmentation. The proposed model utilizes edge templates to extract edge features of an image, and builds edge prior constraints on local regions. Space constraints among local regions of the image are used to express local Gaussian statistical features of the image, and the Gaussian parameters are estimated by maximizing expectations. Then a new local adaptive neighborhood information Gaussian mixture model (GMM) is constructed and an algorithm is proposed to estimate its parameters. The region MRF model that preserves image edge is then built based on the Bayesian theory. The region belief propagation algorithm is applied to globally optimize the IEFRMRF model. Local statistical features are transferred to the image of the global, and image segmentation labels are estimated by MAP criterion during the optimization. Experiments on an artificial noise image and comparisons with the classical Gaussian MRF model and the local region Gaussian MRF model show that the IEFRMRF model not only increases segmentation accuracy rate by 47.9% and 21.4%, respectively, but also acquires sharper edge of segmentation result. The validity of the proposed model is also verified by natural image segmentation experiments.

引用

页码：14 / 19

页数：5

共 15 条

[1] Yao T., Xie Z., Top-down inference with relabeling and mapping rules in hierarchical MRF for image segmentation, Acta Automatica Sinica, 38, 9, pp. 1-13, (2012)
[2] Chen A.Y.C., Corso J.J., Wang L., HOPS: efficient region labeling using higher order proxy neighborhoods, Proceedings of International Conference on Pattern Recognition, pp. 1-4, (2008)
[3] Xu S., Han J., Zhao L., Et al., Algorithm of minimizing local region energy for image segmentation, Journal of Xi'an Jiaotong University, 45, 8, pp. 7-12, (2011)
[4] Gao Q., Roth S., How well do filter-based MRFs model natural images?, Proceedings of DAGM/OAGM Symposium, pp. 62-72, (2012)
[5] Zhang Z., Wang R., Nature image restoration based on edge preserving FoE model, Progress in Natural Science, 19, 9, pp. 1004-1013, (2009)
[6] Koray K., Ercan E.K., Bulent S., Bayesian separation of images modeled with MRF using MCMC, IEEE Transactions Image Process, 18, 5, pp. 982-994, (2009)
[7] Zhang H., Zhang Y., Li H., Et al., Generative Bayesian image super resolution with natural image prior, IEEE Transactions on Image Processing, 21, 9, pp. 4054-4066, (2012)
[8] Katsuki T., Torii A., Inoue M., Posterior mean super-resolution with a causal Gaussian Markov random field prior, IEEE Transactions on Image Processing, 21, 7, pp. 3182-3193, (2012)
[9] Li S.Z., Markov Random Field Modeling in Computer Vision, (2001)
[10] Leung S., Liang G., Solna K., Et al., Expectation-maximization algorithm with local adaptivity, SIAM Journal on Imaging Sciences, 2, 3, pp. 834-857, (2009)

← 1 2 →