Decomposition of digital clumps into convex parts by contour tracing and labelling
Pattern Recognition Letters
| Hi-index | 0.00 |
Aquatic interfaces in the environment are colonized by a large variety of pro- and eucaryotic microorganisms, which may be examined by confocal laser scanning microscopy. We describe an algorithm to identify and count the organisms in multi-channel volumetric datasets. Our approach is an intermediate-level segmentation combining a voxel-based classification with lowlevel shape characteristics (convexity). Local intensity maxima are used as seed points for a watershed transform. Subsequently, we solve the problem of over-segmentation by merging regions. The merge criterion is taking the depth of the 'valley' between adjacent segments into account. The method allows to make correct segmentation decisions without the use of additional shape information. Also this method provides a good basis for further analysis steps, e.g. to recognize organisms that consist of multiple parts.