基于SVDD的交互式区域增长图像分割算法

Interactive region growing for image segmentation based on SVDD

为克服经典区域增长算法中生长规则选取的困难,提出基于交互式支持向量域描述的区域增长图像分割算法。首先通过交互方式选择目标区域的一个种子点,然后通过交互式选择属于该目标区域的子块和非目标区域的子块构造SVDD的训练样本;然后利用这些已知的训练样本训练支持向量域分类器;在区域增长过程中,利用训练所得的SVDD建立增长规则。仿真实验表明,提出的算法是合理可行的。

To solve the difficulty of feature selection and region growing rule existing in conventional region growing image segmentation algorithm, a region growing algorithm based on interactive support vector domain description classifier is proposed. An initial seed is selected by interactive manual operation. At the same time the blocks belonging to the object region or the background region are interactively selected. Then the support vector domain description classifier is trained by the training data collected in the first step. The region growing rules are established by exploiting the support vector domain description classifier in the end. Experimental results show that this algorithm is feasible.