【摘要】 提出了一种基于SVM后验概率的MRF分割方法,将支持向量机的后验概率应用于Markov随机场方法中,通过贝叶斯公式将对样本条件概率的估计转换为后验概率估计,再通过对SVM决策函数输出的映射来产生后验概率,并将SVM估计的后验概率信息带入MRF模型实现分割,从而完成了一种新的Markov随机场模型的分割方法。实验结果表明,采用此方法分割纹理图像可以获得较好的分割结果。更多还原

【Abstract】 A novel segmentation method based on Markov Random Field(MRF) and Support Vector Machine(SVM) posteriori probability is proposed in the paper.As a rule,image segmentation using MRF model has two steps.Firstly,distribution of conditional probability of pixel characteristic is obtained by the parameter estimate for probability density and then maximum a posteriori(MAP) principle is always used to gain the optimum estimate of class label.In practice,the hypothesis of Gauss distribution model is always adopted,but it is not the model fit for any images,for example,SAR images often fit to a model of Rayleigh distribution and especially some texture images,it is very difficult to deduce an accurate distribution model.In order to solve the two major problems which are the complexity of parameter estimate in using the distribution of conditional probability and the difficulty of deducing an accurate distribution in theoretical way,the new segmentation approach based on MRF and SVM posteriori probability is proposed.Support Vector Machine is a set of related supervised learning method,it is a classification technique based on the structural risk minimization principle and it maps input vectors to a higher dimensional space where maximal separating two parallel hyperplanes are constructed.An assumption is made that the larger the margin or distance between these parallel hyperplanes the better the generalisation error of the classifier will be.However,in the pattern recognition practice,people need soft decision,that is to say,not only gain class label which the sample belongs to,but also obtain the membership degree of sample in each class label,that is posteriori probability of sample.The new segmentation algorithm proposed by the paper follows three steps.Firstly,the paper adopts the Platt’s method to obtain the posteriori probability by mapping the output of SVM decision-function after training.Secondly,it converts the conditional probability estimate into posteriori probability estimate in terms of Bayes formula,and then proposes a new segmentation method which depends on MRF model based on posteriori probability.Finally,it brings the information of posteriori estimate into MRF model,thus the posteriori probability based on SVM is combined with MRF in the application of image segmentation.Two experiments has been conducted,one is that the paper selects twelve texture images from Brodatz standard texture database to make up some merged texture images.The number of training samples in each texture class is 30 and eight features are used to characterize each texture sample.The results of three group segmentation experiments all show that the new method appears preferable to Gaussian MRF method.The other experiment is that the synthesis texture image is composed by SAR texture images,SAR images do not fit for Gauss distribution,as a result,Gaussian MRF method results in high misclassification rate and low robustness.On the contrary,the proposed algorithm depends on the information of posteriori probability estimate based on SVM without the hypothesis of sample conditional probability,so it achieves a higher level of robustness and segmentation results demonstrating its efficiency.更多还原