Abstract: Spectral clustering algorithm has been widely used in image segmentation, however, traditional spectral clustering algorithms are susceptible to the effect of color image size and similarity measure. They create large amount of calculation and the poor segmentation result. In order to solve these two problems, this paper proposes a new spectral clustering segmentation algorithm based on geodesic characteristics of super pixel set. It uses the pre-segmentation to get the super set of pixels. Then it constructs a weighted graph, and uses the geodesic distance and color features to build a weight matrix. It uses NJW（Ng-Jordan-Weiss） algorithm to get the segmentation result. The experimental results show that the segmentation accuracy and computational complexity of the algorithm in this paper are improved substantially.

Key words: pre-segmentation, super pixel sets, geodesic distance, Ng-Jordan-Weiss（NJW） algorithm

摘要： 在图像分割中谱聚类算法得到了广泛的应用，但传统谱聚类算法易受到彩色图像大小和相似性测度的影响，导致计算量大和分割精度低的问题。为了解决这两个问题，提出一种新的基于超像素集测地线特征的谱聚类分割算法。该方法通过对彩色图像进行预分割得到超像素集，并以超像素集为基础构造加权图，利用测地线距离特征和颜色特征构造权值矩阵，最后应用NJW（Ng-Jordan-Weiss）算法得到最终的分割结果。对比实验结果表明该算法在分割精度和计算复杂度上都有较大改善。

关键词: 预分割, 超像素集, 测地线距离, Ng-Jordan-Weiss（NJW）算法