Classifications Modification Based FCM with Spatial Information for Image Segmentation

Image segmentation is the foundation of computer vision and pattern recognition but is still a challenging problem. FCM has been used in image segmentation many years but its problem that is sensitive to noise still cannot be solved effectively. This paper proposes an improved algorithm that incorporates the spatial information and classification modification into the fuzzy C-means (FCM), which can overcome this shortcoming of the traditional FCM without increasing the computational complexity. In this algorithm, FCM will be used for initial segmentation, and then the classification of the element will be modified according to the dispersion of the class. After that, selective smoothing will be executed, which changes the gray value of the pixel according to the class its neighboring elements belong to. Finally, the image treated will be segregated again using FCM. As the experimental results shows, this algorithm can suppress noise efficiently and even can be used in image denoising. Moreover, the segmentation result of the uneven illumination image is also nice.