Orientation based segmentation for phase-contrast microscopic image of confluent cell

In this research, we propose a novel segmentation method for image of cultured cell at a confluent state, obtained by phase-contrast microscope, based on the orientation. First, we assign to each pixel in the image the direction of an eigenvector corresponding to a smaller eigenvalue of the 2 by 2 Hessian matrix with respect to brightness. Next, we define the orientation at a certain pixel as the histograms of the direction at pixels in the surrounding regions. Then, we evaluate deviation of histograms in the individual regions by entropy, and regard the series of entropy as a multi-dimensional vector, the dimension of which corresponds with the number of regions. We suppose that the vector is assigned to the pixel of interest. Finally, we segment the image based on the multi-dimensional vector using K-means method. We investigate the efficacy of the proposed method using an actual human confluent fibroblast image acquired by phase-contrast microscopy.