Robust cell segmentation based on gradient detection, Gabor filtering and morphological erosion

Abstract

Research on cell segmentation is experiencing growing pains that need to be addressed for developing generic and robust techniques. The diversity of cells is growing rapidly and the demand for segmenting different types of cell images continue to increase in recent years. It is still very challenging for existing methods to segment various types of cells automatically and robustly. In this paper, we try to address this challenge by proposing an approach that is capable of segmenting various types of cells robustly. To avoid the effect of the global intensity variations, the gradients of cells are computed and then smoothed by the Gabor filter to generate the gradient image with increased intensity uniformity. To find the optimal method for cell foreground segmentation, we evaluated and compared state of the art threshold selection methods extensively. The slope difference distribution (SDD) method was testified as the optimal threshold selection method for cell segmentation and its optimal parameters were obtained in this paper based on a variety of cell images. We tested several morphological erosion methods and combined the iterative erosion method and the area-constrained ultimate erosion method to separate the connected cells robustly. Thirteen types of cells were used to compare the proposed approach with state of the art approaches. The quantitative comparison showed that the proposed approach achieved the highest average F-measure accuracy, 95.61 %. Thus, the proposed approach has the potential to be useful for various microscopic applications.