Article Preview
TopIntroduction
Image processing includes several subdomains like image segmentation, filter application, image noise reduction, etc. Image segmentation is a step on which several steps in image processing are based. It consists of grouping the pixels of an image into groups with common characteristics, such as grayscale, color, texture, etc. Its importance lies in the fact that the results of subsequent steps in image processing depending on the quality of this segmentation.
In this paper, the authors are interested in improving an image segmentation method, named Fuzzy c-means (FCM). Region-based (Ciesielski &Udupa, 2010) (Nakib, Oulhadj, & Siarry, 2009) and edge-detection-based approaches (Papari & Petkov, 2011) are the most used techniques in image segmentation. FCM is an unsupervised classification method, which is often used in image segmentation. FCM starts with the number of clusters “C” that is set manually (a wrong choice of cluster number can negatively influence the final results). The centers of clusters are initialized randomly. From these centers the membership degree of each pixel is calculated and updated, after that, the centers (Mobile Center Methods) are updated. In this work, the authors are interested in solving two major problems of FCM (Class centers initialization and the fittest number of classes). The authors propose two approaches. In the first one, the authors find the near-optimal cluster centers (using metaheuristics), in the second one, the authors determine automatically the number of clusters (with statistical formulas).