基于全局阈值分割和局部聚类的自然图像超像素分割技术研究

项目名称： 基于全局阈值分割和局部聚类的自然图像超像素分割技术研究

项目编号： No.61502065

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 计算机科学学科

项目作者： 龙建武

作者单位： 重庆理工大学

项目金额： 20万元

中文摘要： 随着自然图像分辨率越来越高，以致实时图像处理面临着一个新的挑战，而用图像区域取代像素点的超像素分割技术可以大幅度消减图像冗余信息，从而简化图像内容并为后续实时图像处理奠定了基础。至此，超像素分割技术现已成为图像处理和计算机视觉领域中的一个研究热点。本项目针对现有方法存在的实时性较差、过分割现象较为严重、分割误差较大等问题，将对超像素分割技术展开深入研究。首先利用全局阈值算法的实时分割性能，拟提出一种基于直方图的自然图像快速多级阈值不规则超像素分割算法；然后以多级阈值分割为基础，结合规则网格划分方式拟建立规则超像素种子点快速定位方案；最后基于多种子点和种子区域生长算法，拟提出一种多特征融合的规则超像素局部聚类算法。本项目为自然图像不规则/规则超像素分割技术的研究开辟了新的思路，为后续图像分析、处理、理解、识别等领域的应用和研究提供了理论基础和实时性保障，具有非常重要的研究意义。

中文关键词： 超像素；超像素分割；图像分割；全局阈值分割；局部聚类

英文摘要： With growing of the resolution of natural images, it is a challenge task for real-time image processing. But superpixel segmentation technique is alternative solution based on image regions instead of pixels greatly reducing image redundancy and simplifying image content, which builds a solid foundation for the subsequent real-time image processing. Nowadays the superpixel segmentation technology is widely concerned in the fields of image processing and computer vision. This project launches a deep research for the superpixel segmentation techniques considering expensive computation, seriously over-segmentation and poor segmentation performance of the existing algorithms. Firstly we propose a fast histogram-based multilevel thresholding irregular superpixel segmentation algorithm for natural images taking fully account of the real-time performance of global thresholding approaches. Secondly we establish a fast regular superpixel-seeds locating scheme combining regular grid partitioning with multilevel thresholding. Thirdly we present a regular superpixel local clustering algorithm fusing multi features based on multi seeds and seeded region growing. This research opens a new train of thought for natural images irregular and regular superpixel segmentation techniques, which provides theoretical basis and real-time support for the subsequent application and research in the fields of image parsing, processing, understanding, recognition etc., and is of very huge researching significance.

英文关键词： Superpixel;Superpixel Segmentation;Image Segmentation;Global Thresholding;Local Clustering