Subimage Cosegmentation in a Single White Blood Cell Image

As one of the most basic and key steps in the white blood cell (WBC) automatic recognition system, the accuracy and stability of WBC segmentation greatly affect the recognition accuracy of the whole system. Inspired by co segmentation, this paper presents a new method to obtain the entire WBC contour in order to solve the adhesion problem. Firstly, the color space transformation and thresholding are employed to obtain sub images. Among them, a sub image called reference image only includes plasma and red blood cells (RBCs), while the others contain WBCs. Then, a similarity measurement is introduced into the region-based active contour model in order to co segment between the reference image and the other sub images. Due to full use of cross information, such as color and shape, the proposed method has been successfully applied to a large number of microscopic cell images, showing promising segmentation results for diverse cell appearance and image quality.