Image Sequence Segmentation Combining Global Labeling and Local Relabeling and its Application to Materials Science Images Jarrell W. Waggoner a , Jeff Simmons b , and Song Wang a a University of South Carolina, Columbia, SC 29208, USA; b Materials and Manufacturing Directorate, Air Force Research Labs, Dayton, OH 45433, USA ABSTRACT Accurately segmenting a series of 2D serial-sectioned images for multiple, contiguous 3D structures has important applications in medical image processing, video sequence analysis, and materials science image segmentation. While 2D structure topology is largely consistent across consecutive serial sections, it may vary locally because a 3D structure of interest may not span the entire 2D sequence. In this paper, we develop a new approach to address this challenging problem by considering both the global consistency and possible local inconsistency of the 2D structural topology. In this approach, we repeatedly propagate a 2D segmentation from one slice to another, and we formulate each step of this propagation as an optimal labeling problem that can be efficiently solved using the graph-cut algorithm. Specifically, we divide the optimal labeling into two steps: a global labeling that enforces topology consistency, and a local labeling that identifies possible topology inconsistency. We justify the effectiveness of the proposed approach by using it to segment a sequence of serial-section microscopic images of an alloy widely used in material sciences and compare its performance against several existing image segmentation methods. Keywords: Segmentation, Materials, Propagation, Topology Constraints, Local and Global 1. INTRODUCTION Images of 3D structures made up of multiple 2D slices play an important role in a myriad of fields, including video analysis and compression, 1 medical imaging, 2 and civil and industrial materials science. 3 Everything from tomographic sequences and 3D structure volumes to medical CT/MRI and video sequences make up this vast array of serial-sectioned data, which form a collectively challenging set of problems for image segmentation. While significant research has been made on many of these problems, one basic issue has not been specifically and systematically addressed: how to model and identify both 2D topology consistency and possible inconsistency across slices in image sequence segmentation. In this paper, we address this issue by developing an image sequence segmentation method, which propagates a 2D segmentation sequentially from one slice to another. As illustrated in Fig. 1, the 3D structure of interest is made up of multiple contiguous substructures, and the 2D topology 4, 5 consistency is reflected by the fact that nearby series sections show similar substructures (e.g., s 1 ↔ s 2 ). However, inconsistency may be introduced when the series section moves into a new substructure or moves out of an existing substructure (e.g., s 2 ↔ s 3 , and s 3 ↔ s 4 ). The segmentation of such structures becomes a very challenging problem when the number of substructures is large and such 2D topology changes are not known. This is a common phenomenon in many fields, such as segmenting cells in medical imaging, grain structures in materials science, and crowd scenes in video surveillance. Some previous work employs direct 3D segmentation instead of segmenting 2D slices sequentially; however, direct 3D segmentation may not work when there is large intensity and contrast changes across these slices 6 and/or inter-slice resolution is much lower than the intra-slice resolution. 7 The method proposed in this paper specificially focuses on segmenting images with a large number of contiguous substructures. Further author information: (Send correspondence to J.W.W.) J.W.W.: E-mail: waggonej@email.sc.edu, Telephone: 847-261-4747 J.S.: E-mail: jeff.simmons@wpafb.af.mil S.W.: E-mail: songwang@cec.sc.edu, Telephone: 803-777-2487 Computational Imaging X, edited by Charles A. Bouman, Ilya Pollak, Patrick J. Wolfe, Proc. of SPIE-IS&T Electronic Imaging, SPIE Vol. 8296, 829606 © 2012 SPIE-IS&T · CCC code: 0277-786X/12/$18 · doi: 10.1117/12.906471 SPIE-IS&T/ Vol. 8296 829606-1 Downloaded from SPIE Digital Library on 27 Feb 2012 to 128.46.115.219. Terms of Use: http://spiedl.org/terms