A SigmaIFCM Algorithm For Image Segmentation In Brain Magnetic Resonance Images

Image segmentation is an important process in image analysis, also is a very difficult process. Currently there is no all-purpose image segmentation algorithm which can be applied in different conditions, and so many people do much research in this field. Fuzzy c-means (FCM) algorithm is a popular image segmentation algorithm among medical image processing field, but it also has many shortages, especially for images with much noise. Magnetic Resonance (MR) image is widely used in medical image analysis and always have much noise in it, so the question of brain MR image segmentation is a very popular topic for research.The data set in FCM algorithm is all pixels in image, and is very huge. Gray level can be used for the data set since gray value of pixel is the calculating basis of membership of pixel in the iterating process of FCM algorithm. So membership of gray level can be gotten when iterating process is finished, then membership of each pixel also can be available with correspondence of gray level and pixel. Besides, the iterating times of FCM algorithm can be reduced using the result cluster centroids of HCM algorithm as the initial cluster centroids of FCM algorithm, and the CPU time can be cut down using the idea of modifying membership value in iterating process.FCM algorithm is not so applicable for medical image with much noise. The membership of neighbor pixels can be considered in the process of calculating membership of the pixel to remove the influence of noise for image segmentation result. Sigma filter idea can be applied for defining of neighbor pixels to preserve edge characteristic of cluster, and the segmented image can be processed using theory of burr-removing and edge-smoothing to correct some wrong-segmented pixels."Bench mark"idea is very important for evaluation of image segmentation algorithms. Mcgill University brainWeb Simulated Brain Database provides an image process evaluation standard which is very popular in image process field. This evaluation method is used to evaluate various image segmentation algorithms to get a factual comment. Since the standard segmentation result is know in advance, this method can be applied to get a quantity evaluation for FCM and its improved algorithms. It's also available for any other image segmentation algorithms.