A Run-Based One-Scan Labeling Algorithm

  • Authors:

  • Lifeng He;Yuyan Chao;Kenji Suzuki;Hidenori Itoh

  • Affiliations:

  • The Faculty of Electronical and Information Engineering Shannxi University of Science and Technology, Shannxi 710021, China The Graduate School of Information Science and Technology, Aichi Prefect ...;The Faculty of Mechanicial and Electronical Engineering Shannxi University of Science and Technology, Shannxi 710021, China The Graduate School of Environment Management, Nagoya Sangyo University, ...;The Department of Radiology, Division of the Biological Sciences, The University of Chicago, Chicago, USA 60637;The Department of Information Engeering, Nagoya Institute of Technology, Showa-ku, Japan 466-8666

  • Venue:
  • ICIAR '09 Proceedings of the 6th International Conference on Image Analysis and Recognition
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents a run-based one-scan algorithm for labeling connected components in a binary image. Our algorithm is different with conventional raster-scan label-equivalence-based algorithms in two ways: (1) to complete connected component labeling, all conventional label-equivalence-based algorithms scan a whole image two or more times, our algorithm scans a whole image only once; (2) all conventional label-equivalence-based algorithms assign each object pixel a provisional label in the first scan and rewrite it in later scans, our algorithm does not assign object pixels but runs provisional labels. In the scan, our algorithm records all run data in an image in a one-dimensional array and assigns a provisional label to each run. Any label equivalence between runs is resolved whenever it is found in the scan, where the smallest label is used as their representative label. After the scan finished, all runs that belong to a connected component will hold the same representative label. Then, using the recorded run data, each object pixel of a run is assigned the representative label corresponding to the run without rewriting the values (i.e., provisional labels) of object pixels and scanning any background pixel again. Experimental results demonstrate that our algorithm is extremely efficient on images with long runs or small number of object pixels.