Semantic segmentation for binarized image using minimal cross section

Tomofumi Narita, Takashi Ijiri

Research output: Contribution to journalArticlepeer-review


This paper presents a semantic segmentation technique for three dimensional (3D) X-ray Computed Tomography (CT) images of natural objects, such as insects or plants. Our technique is based on knowledge that joints of semantically different parts of natural objects are often narrow. Given a binarized 3D CT image, we recursively detect the narrowest cross section that divides the foreground region into two parts. Our narrowest cross section detection consists of three steps; (i) splitting the foreground by erosion operations, (ii) regrowing the split regions by dilation operation and (iii) finding the narrowest cross section in the dilated region by adopting a graph cut method. To evaluate the accuracy of our technique, we adopt it to artificially generated images and found that error pixel rate was less than 2 %. To illustrate the feasibility of our technique, we adopt it to 3D CT images of insects and plants. As results, our technique successfully segmented multiple florets from an inflorescence, stems of a succulent plant, and legs of insects.

Original languageEnglish
JournalJournal of the Institute of Image Electronics Engineers of Japan
Issue number4
Publication statusPublished - 2018


  • Graph cut algorithm
  • Morphological operations
  • Narrow cross section
  • Semantic segmentation
  • X-ray computed tomography images

ASJC Scopus subject areas

  • Computer Science (miscellaneous)
  • Electrical and Electronic Engineering


Dive into the research topics of 'Semantic segmentation for binarized image using minimal cross section'. Together they form a unique fingerprint.

Cite this