The Image Segmentation Based On K-Means With MAP

A major feature of quantitative cell biology is the identification of various cell compartments, cell types and their relationship. Automating of this problem has proven non-trivial, however, and it involves the object of multi-class image partition tasks that are difficult due to the high similarity of objects from various groups and irregularly found structures. To overcome this problem purpose, we propose k-means image segmentation method. And also in the current implementation of the proposed algorithm, the overall segmentation performance of the method can be confined by the graph generation quality. So the future work can be the development of a Maximum Posterior (MAP) estimation for graph generation that optimizes the graph structure jointly with label inference. On the other hand, it is valid to mentioning that the small margin of improvement by the proposed graph based spliting over segnet is because features learned by the CNN are minimizing the cost function rather than the cost function of the polytree.