Automatic Segmentation of Chromosome Cluster Images Into Individual Chromosomes.

Human observers can easily recognize objects having some common features even though their shapes and their environment may appear quite different. In order for a computer to perform a similar task, a model of the class of objects in the form of one or more computer algorithms must be devised. In this research, algorithms are developed to separate two or more touching or overlapping chromosomes by quantifying the single chromosome characteristics such as the variation of the direction angle along the boundary of the chromosome, and the occurrence of straight segments, curved ends and parallel sides along the boundary. The first part of the overall procedure developed in this work recognizes the various features of the chromosome cluster boundary such as straight segments, curved ends, convex bends, concave bends, bumps and parallel sides. The second part takes the results of the first part, as summarized in a matrix called the straight segment matrix, and performs the segmentation, i.e. separation into individual chromosomes, by recursively splitting the initial straight segment matrix into submatrices until only matrices representing single chromosomes remain.