Modified localized multiplicative graph cuts based active contour model for object segmentation based on dynamic narrow band scheme. (March 2017)

• Record Type: Journal Article
• Authors: Zheng, Qiang; Li, Honglun; Fan, Baode; Wu, Shuanhu; Xu, Jindong; Cao, Zhulou
• Is Part Of: Biomedical signal processing and control. Volume 33(2017)
• Page Start: 119
• Page End: 131
• Publication Date: 2017-03
• Subjects: Active contour model -- Graph cuts -- Image segmentation; Signal processing -- Periodicals; Biomedical engineering -- Periodicals; Signal Processing, Computer-Assisted -- Periodicals; Image Processing, Computer-Assisted -- Periodicals; Biomedical Engineering -- Periodicals; 610.28
• DOI: 10.1016/j.bspc.2016.11.019
• Languages: English
• ISSNs: 1746-8094

Abstract

Highlights: Static narrow band in existing LM-GCACM makes segmentation inconvenient. Dynamic narrow band allows initial curve be any size or shape inside object. Novel GC formulation for region term with contrast constraint and self-constraint. Abstract: Localized multiplicative graph cuts based active contour model (LM-GCACM) has been widely utilized in object segmentation. However, the curve evolution of existing LM-GCACMs is based on static narrow band scheme generally, which is inconvenient in object segmentation because it requires the initialized curve be close to object boundary, and the narrow band is difficult to be determined. In this paper, a modified LM-GCACM based on dynamic narrow band is proposed to improve static narrow band. The dynamic narrow band allows the initialized curve to be any size or shape as long as it is inside object, and the narrow band can be built between the evolving curve and image bounding box. There are three contributions made to achieve dynamic narrow band. Firstly, the multiplicative region term is modified more suitable for segmentation. Secondly, a contrast constraint term is introduced to help evolving curve to go over false edges in the curve inflation evolution process. Thirdly, a self-constraint term is proposed to reduce the influence of surrounding clutter around object in the background, and guarantee segmentation stop on object boundary. Experiments on synthetic and medical images demonstrate the advantages of the proposed method.