1 Department of Clinical Medicine - Stereological Research Laboratory, Department of Clinical Medicine, Health, Aarhus University2 Department of Clinical Medicine - Stereological Research Laboratory, Department of Clinical Medicine, Health, Aarhus University
As the use of digital images have become standard in everyday bioimaging, people are naturally trying to utilize computer algorithms for automating laborious and repetetive image analysis tasks. A preliminary task is the segmentation of structures of interest from digital images. This may in itself be a major task, but in bioimaging and tissue quantification it is often complicated further by the need for segmenting images of overlapping particles, for instance neurons. One approach to segmenting overlapping particles is to oversegment the image into many small regions which are then combined into the correct shapes in a postprocessing step. The postprocessing step is unfortunately often both difficult and computationally expensive. Another approach is to incorporate descriptions of the overlapping shapes into a segmentation algorithm which normally only segments the union of all particle profiles. This may, however, quickly lead to the implementation of complex descriptions of any possible configuration the overlapping shapes may appear in. Presented here is a new approach to segment overlapping shapes which utilizes information gained from probing the image with test rays. Test rays intersections provide exact information about shape contours which can be exploited for the separation of overlapping shapes. The resulting procedure may be performed either as a separate postprocessing step after a single contour of all particles is segmented or it may be performed in combination with an existing segmentation algorithm. Furthermore, the approach handles shapes of different size and also both convex and non-convex shapes. Results and performance on simulated images will be shown.