EVENT
Event News
Talks by Dr. Wan: "Talk 1: Graph Cut and Linear Algebra Approach to Cell Image Segmentation"
Date/Time:
July 25th (Tuesday), 11:00-12:00am
Place:
National Institute of Informatics
Room: 12F, 1208
Title:
Graph Cut and Linear Algebra Approach to Cell Image Segmentation
Speaker:
Dr. Justin W.L. Wan
Canada Research Chair in Scientific Computing,
Associate Professor, SciCom group in the David R. Cheriton School of Computer Science at University of Waterloo, Canada,
Director of the Centre for Computational Mathematics in Industry and Commerce (CCMIC).
https://cs.uwaterloo.ca/~jwlwan/
Abstract:
Segmentation of cells in time-lapse bright-field microscopic images is crucial in understanding cell behaviors for medical research. However, the complex nature of the cells, together with poor contrast, broken cell boundaries and the halo artifact, pose nontrivial challenges to this problem. In this talk, we present robust mathematical models based on linear algebra techniques to segment bright-field cells automatically. One approach is to formulate image segmentation as graph cut problems. We combine the techniques of graph cut, multiresolution, and Bhattacharyya measure, performed in a multiscale framework, to locate multiple cells in bright-field images. Another approach is to treat cell image segmentation as a background subtraction problem. It can be formulated as a robust Principal Component Pursuit (PCP) problem which minimizes the rank of the image matrix. In this approach, we exploit the sparse component of the PCP solution for identifying the cell pixels. However, the sparse component and the nonzero entries can scatter all over the image, resulting in a noisy segmentation. We improve the model by combining PCP with spectral clustering. Spectral clustering makes use of the eigenvectors of the graph Laplacian matrix to classify data. Seemingly unrelated approaches, we combine the two techniques by incorporating normalized-cut in the PCP as a measure for the quality of the segmentation. Experimental results demonstrate that the proposed models are effective in segmenting cells obtained from bright-field images.
Contact:
Ken Hayami (hayami(at)nii.ac.jp)