基于结构张量特征值的纹理图像分割模型

Texture Image Segmentation Model Based on Eigenvalues of Structure Tensor

通过对结构张量的研究和对纹理图像的分析,提出了一种基于结构张量特征值的标量型纹理特征描述,将其和原图像分别嵌入到两相模糊区域竞争模型和CV模型中,给出了一种纹理和灰度相结合的无监督纹理图像分割模型.为获得新模型的全局最优解,采用了Chambolle对偶法加以实现.针对自然和合成纹理图像进行了相关实验,结果表明该模型特征数据维数少,具有较快的收敛速度和更准确的分割效果.

Through the research on the structure tensor and the analysis of the texture image,a scalar texture feature descriptor based on eigenvalues of structure tensor is designed.Embedding the descriptor and the original image into the two-phase fuzzy region competition model and the CV model respectively,an unsupervised texture image segmentation model which combines texture with intensity information is given.The Chambolle dual method is adopted in order to minimize the optimum global solution of the new model.According to the related experiments for natural and synthetic images,the new method has some advantages including faster convergence speed due to less feature channels and more precise segmentation results.