结合半局部信息与结构张量的无监督纹理图像分割

Unsupervised texture segmentation combined semi-local image information and structure tensor

针对纹理的半局部性质与方向性,提出一种基于活动轮廓的无监督双纹理图像分割方法。首先分析基于半局部信息的纹理特征表示方法,指出其不能有效区分纹理的方向特征。然后利用半局部信息结合非线性结构张量构造4通道纹理特征,采用混合高斯模型作为纹理特征的概率密度函数,模型的数值求解采用分裂Bregman方法。新方法充分考虑纹理结构的周期性与方向特性,实验结果表明,其能够处理复杂的双纹理图像分割任务,同时具有高效与无监督特性。

An unsupervised two-phase texture segmentation method based on active contour was proposed. First, the texture feature extraction approach based on semi-local image information was analyzed, which revealed that it could not represent texture＇ s orientation information. In order to segment texture images containing periodic and orientational character, a fourchannel texture feature was achieved combing semi-local image information with nonlinear structure tensor. Then Gaussian mixture model was adopted to describe the probability density function of the features. Numerical algorithms were based on split Bregman method. Experimental results for both nature and synthetic texture images show that our method could cope with complex segmentation tasks. Meanwhile it is effective and unsupervised.