A new approach to separating touching spots in particle images

In particle images, we call objects which touch or overlap touching spots. Separating touching spots into single ones is an important step in many cases. Recently, several separating algorithms have been proposed, all based on obvious features, such as sharp concavity or strong gray-contrast between two spots. However, touching spots without such features are difficult to separate when using existing algorithms, for example, touching cells in bacteria images where there are different sizes of spots with weak contrast. We propose a new algorithm, based on 8-connected boundary tracking, for separating these special touching spots. The outer shell of a spot is first tracked and then stripped gradually. If the tracking bug meets a point on the spot boundary, then there may be two touching spots. The length between the tracking point and the current point along the boundary and also the length of the whole boundary perimeter are computed. If the two lengths are larger than a threshold, the touching spots have to be separated into two or more single spots. Our algorithm is different from those based on morphological erosion and distance transform, and it does not require to have an acute angle and/or a distinct grey-contrast at points where spots touch. We have developed a software system for separating the touching cells in a bacteria image using our algorithm. Experimental results from our algorithm show that it can effectively separate touching cells in a bacteria image and it is also shown that 10% more touching cells are separated than using other algorithms.