Abstract:

The segmentation result of the traditional region growing algorithm depends on the selection of the seed point. The noise of the image and the uneven grayscale value are easy to form the segmentation cavity in the process of segmentation. Aiming at the above problems, an improved region growing algorithm based on superpixel is proposed. Frist of all, the Laplacian sharpening is used to enhance the boundary of the target to be segmented. According to the features of gray similarity, the SLIC（Simple Linear Iterative Clustering） superpixel segmentation method is used to segment the original image into several irregular regions. Then an undirected weighted graph based on irregular regions will be established. A region is selected as a seed, the region is grown in units of the segmented irregular regions according to the undirected weighting map. To clarify the edge area, the region growing algorithm in pixels runs at the edge of the segmentation target finally. Compared with the traditional region growing algorithm, the improved algorithm is less affected by the seed point selection in the segmentation result, and the improved algorithm can effectively solve the problem of segmentation holes. Compared with clustering segmentation, Otsu threshold segmentation method, the proposed algorithm has obvious advantages in segmentation accuracy.

Key words: Laplacian, Simple Linear Iterative Clustering（SLIC）, regionl growing, target segmentation

摘要：

传统区域生长算法的分割结果依赖于种子点的选取，且图像自身的噪声以及灰度值不均匀等问题易在分割目标过程中形成分割空洞，针对以上问题提出了基于超像素的改进区域生长算法。采用拉普拉斯锐化，增强待分割目标边界，之后根据像素灰度相似的特征采用SLIC（简单线性迭代聚类算法）超像素分割将原始图像分割成若干不规则区域，建立不规则区域间的无向加权图，选取种子区域，根据无向加权图以分割好的不规则区域为单位进行区域生长，最后在分割目标边缘处以像素为单位做区域生长，细化边界。对比于传统区域生长算法，改进后的算法在分割结果上受种子点选取影响较小，且能有效地解决分割空洞等问题。对比于聚类分割，Otsu（最大类间方差）阈值分割法等典型算法，该算法在分割精度上具有明显优势。

关键词: 拉普拉斯锐化, 简单线性迭代聚类算法（SLIC）, 区域生长, 目标分割