一种新的图像超像素分割方法

针对现有超像素分割方法无法自动确定合适的超像素数目，以及难以有效贴合图像目标边界等问题，该文提出一种新的利用局部信息进行多层级简单线性迭代聚类的图像超像素分割方法。首先，运用基于局部信息的简单线性迭代聚类(LI-SLIC)对原始图像进行超像素初分割，然后，根据超像素的色彩标准差对其进行自适应多层级迭代分割，直至每个超像素块的色彩标准差小于预设阈值，最后，利用相邻超像素间的色彩差异对过分割的超像素进行合并。为验证方法的有效性，该文采用Berkeley, Pascal VOC和3Dircadb公共数据库作为实验数据集，并与其他多种超像素分割方法进行了比较。实验结果表明，该文提出的超像素分割方法能更精确贴合图像目标边界，有效抑制图像过分割和欠分割。

A New Method for Image Superpixel Segmentation

Considering the problem that the existing superpixel methods are usually unable to set an appropriate number of generated superpixels automatically and unable to adhere to image boundaries effectively, a new superpixel method is proposed in this paper, which utilizes local information to perform multi-level simple linear iterative clustering to generate superpixels. First, original image is initially segmented by Simple Liner Iterative Clustering based on Local Information (LI-SLIC). Then, each superpixel is segmented iteratively until its color standard deviation is lower than a predefined threshold. Finally, the over-segmented superpixels are merged based on the color differences between adjacent superpixels. Experiments on Berkeley, Pascal VOC and 3Dircadb databases, as well as comparison with other methods indicate that the proposed method can adhere to image boundaries more accurately, and can prevent over- and under- segmentations more effectively.