基于纹理特征的多区域水平集图像分割方法

现有多区域水平集方法大多利用复杂的能量函数来驱动多个水平集函数的演变，这样不仅模型复杂且存在很多限制.为此本文提出一种基于纹理特征的多区域水平集方法，利用任意数量的水平集函数来对相应数量的图像区域进行分割.本文首先对图像的颜色和纹理信息建立联合分布并将其代入能量函数；引入平滑概率标签，根据概率性质建立基于标签驱动的多区域水平集迭代更新方程.之后将每个水平集投影到离散概率空间得到一系列近似标签，并由这些标签得到基于多区域水平集的先验概率，从而将多个轮廓演变信息代入统计框架.而不同区域的统计参数也通过最小化能量函数由概率标签迭代更新.通过与其他分割算法在大量复杂实景图像上的实验对比，验证了本文算法的有效性.

A Multi-region Level Set Model Based on Texture Feature for Image Segmentation

Most of the existing level set methods for multi-region use complex energy functions to drive the evolution of multiple level sets,which not only makes the model more complex but also has many limitations.In this paper,we propose a fast multi-region active contour method based on texture features by using an arbitrary number of level sets functions to segment an image into regions of the corresponding amount.We first establish a joint distribution of color and texture information and bring it into data term of energy function.Then,we introduce the smooth probability label and establish an update equation of level set function for multi-region driven by probability labels.And,each level set for different regions is projected into the discrete space to get a series of approximate labels.Due to these labels,a prior probability based on multi-level-sets is obtained,which introduce the evolution information of multiple contours into the statistical framework.The statistical parameters of each region are also updated iteratively from the smooth probability labels by minimizing the energy function.We experimentally compare the proposed approach with other methods on complicated real-world images and demonstrate its good performance in practice.