Fast ISODATA clustering algorithms
Pattern Recognition
Information Retrieval
On candidates selection for hysteresis thresholds in edge detection
Pattern Recognition
Seeded watersheds for combined segmentation and tracking of cells
ICIAP'05 Proceedings of the 13th international conference on Image Analysis and Processing
The Watershed Transform: Definitions, Algorithms and Parallelization Strategies
Fundamenta Informaticae
Bioscientific data processing and modeling
ISoLA'12 Proceedings of the 5th international conference on Leveraging Applications of Formal Methods, Verification and Validation: applications and case studies - Volume Part II
| Hi-index | 0.00 |
High-throughput microscopy imaging applications represent an important research field that is focused on testing and comparing lots of different conditions in living systems. It runs over a limited time-frame and per time step images are generated as output; within the time-range a resilient variation in the images of the experiment is characteristic. Studies represent dynamic circumstances expressed in shape variation of the objects under study. For object extraction, i.e. the segmentation of cells, aforementioned conditions have to be taken into account. Segmentation is used to extract objects from images and from objects features are measured. For high-throughput applications generic segmentation algorithms tend to be suboptimal. Therefore, an algorithm is required that can adapt to a range of variations; i.e. self-adaptation of the segmentation parameters without prior knowledge. In order to prevent measurement bias, the algorithm should be able to assess all inconclusive configurations, e.g. cell clusters. The segmentation method must be accurate and robust so that results that can be trustfully used in further analysis and interpretation. For this study a number of algorithms were evaluated and from the results a new algorithm was developed; the watershed masked clustering algorithm. It consists of three steps: (1) a watershed algorithm is used to establish the coarse location of objects, (2) the threshold is optimized by applying a clustering in each watershed region and (3) each mask is reevaluated on consistency and re-optimized so as to result in consistent segmented objects. The evaluation of our algorithm is realized by testing with images containing artificial objects and real-life microscopy images. The result shows that our algorithm is significantly more accurate, more robust and very reproducible.