Automatic cell region detection by k-means with weighted entropy

In this paper, we propose an automatic method to detect human embryonic stem cell regions. The proposed method utilizes the K-means algorithm with weighted entropy. As in phase contrast images the cell regions have high intensity variation, they usually yield higher entropy values than the substrate regions which have less intensity variation. Thus, the entropy can be used as an important feature for the detection of stem cells. However, homogeneity in intensity within some of the cell bodies and halos surrounding the cell bodies also gives low entropy values. Therefore, we introduce a weighted entropy formulation which fuses entropy and image intensity information to detect the entire cell regions.