北京邮电大学学报

  • EI核心期刊

北京邮电大学学报 ›› 2016, Vol. 39 ›› Issue (5): 51-55.doi: 10.13190/j.jbupt.2016.05.011

• 论文 • 上一篇    下一篇

基于格式塔认知框架的腹腔CT多目标分割算法

冯筠, 刘飞鸿, 李盼, 卜起荣, 王红玉   

  1. 西北大学 信息科学与技术学院, 西安 710127
  • 收稿日期:2016-01-06 出版日期:2016-10-28 发布日期:2016-12-02
  • 作者简介:冯筠(1972-),女,教授,E-mail:fengjun@nwu.edu.cn.
  • 基金资助:
    国家自然科学基金项目(61372046,61272286);陕西省自然科学基金项目(2014JM8338);榆林市产学研合作项目(2012cxy3-5)

Multi-Object Segmentation for Abdominal CT Images Based on Gestalt Cognitive Framework

FENG Jun, LIU Fei-hong, LI Pan, BU Qi-rong, WANG Hong-yu   

  1. School of Information and Technology Northwest University, Xi'an 710127, China
  • Received:2016-01-06 Online:2016-10-28 Published:2016-12-02

摘要: 为了解决腹腔软组织电子计算机断层扫描影像难于分割的难题,提出一种基于格式塔认知框架的多目标分割算法.通过借鉴格式塔认知框架中“邻近度、相似度”的思想,引入超像素算法处理电子计算机断层扫描图像处理,生成可视块.进一步地,在可视块粒度描述有向邻接关系,以软组织的相对空间位置约束聚类分割过程.在公开数据库上的实验结果表明,该算法降低了聚类的计算量,其结果比当前流行的算法准确率更高.

关键词: 格式塔, 腹腔电子计算机断层扫描影像, 分割, 可视块, 分类, 空间相关性

Abstract: In order to acquire better organ segmentation from abdominal computed tomography (CT) images, a multi-object segmentation algorithm based on cognitive framework is proposed. Inspired by the proximity and similarity idea in gestalt, super pixel concept of CT image processing has been produced. Specifically, by establishing the directed adjacency relationship of super-pixel, the spatial relationships of abdominal organs are modeled as prior knowledge to improve the classification accuracy. Experiments in public datasets illustrate that the proposed algorithm achieves better performance in either speed or accuracy than that of the state-of-art methods.

Key words: Gestalt, abdominal computed tomography, segmentation, visual patch, classification, spatial relationship

中图分类号: