基于超像素的快速图像分割-中国科技论文在线.PDFVIP

中国科技论文在线 基于超像素的快速图像分割# 程再兴，马尽文** （北京大学数学科学学院信息科学系和数学及其应用教育部重点实验室，北京100871） 5 摘要：图像分割是图像分析与理解中的一个基本步骤,其目的是将图像按像素划分成若干个 区域，而每个区域与图像中的一个真实物体或背景相对应。本文采用超像素的思想提出了 一种快速的图像分割方法。我们首先采用 SLIC 算法对图像进行一定的过分割而形成超像 素，然后以超像素为基本单元构造无向加权图，最后对超像素进行自适应谱聚类得到一个 10 较为均衡的图像处理分割结果。实验结果表明，这种基于超像素的快速图像分割新方法不 仅拥有很高的运行效率，并且可获得较高水平的分割结果。 关键词：图像分割；超像素；无向加权图；谱聚类；竞争学习 中图分类号：TP391 15 Fast Image Segmentation Through Super-Pixels Analysis CHENG Zaixing, MA Jinwen (Department of Information Science, School of Mathematical Sciences and LMAM, Peking University, Beijing 100871) Abstract : Image segmentaion is a basic step for image analysis and understanding, which tries to 20 divide all the pixels in an image into a number of regions that correspond to the objects and background in the image, respectively. In this paper, we propose a fast image segmentation method with a process of super-pixels analysis. Specifically, we firstly implement the SLIC algorithm on the image for a proper over-segmentation to form a set of super-pixels, and then regard these super-pixels as the basic nodes to construct an undirected weighted graph, and finally 25 utilize an adaptive spectral clustering algorithm to classify the super-pixels as well as the corresponding pixels and obtain a balanced image segmentation. It is demonstrated by the experimental results that this proposed super-pixels based fast image segmentation method not only runs very fast but also keep a relatively good segmentation accuracy as well. Key words: image segematio