Membership Adjusted Superpixel Based Fuzzy C-Means for White Blood Cell Segmentation

Fuzzy C-means (FCM) is a well-known clustering technique that is efficiently used for image segmentation. However, the performance of the FCM degrades for noisy images and slow convergence due to the repeated calculation of the distance among pixels and cluster centers. Therefore, this study develops a Membership Adjusted superpixel-based Fuzzy C-Means (MASFCM) to overcome both issues. The proposed MASFCM utilizes superpixel image as input to make the clustering method fast and noise robust. Membership scaling based on triangle inequality has been employed to speed up performance by avoiding unnecessary distance calculations. Lastly, the final membership matrix has been filtered using morphological reconstruction to enhance the robustness of the MASFRFCM. Furthermore, the proposed MASFCM has been efficiently applied to segment the White Blood Cell (WBC) from pathology images. The visual and numerical results clearly demonstrate that the proposed MASFCM produces promising outcomes compared to other tested state-of-the-art clustering techniques over clean as well as noisy pathology images.