Researching On Image Segmentation Algorithms Based On Fuzzy C-means Clustering

Image segmentation is a key step from image processing to image analysis, and has animportant position in the field of image recognition and computer vision. In recent years,fuzzy clustering is applied to image segmentation algorithms, one of the most famous is thefuzzy c-means (FCM) algorithm. However, FCM algorithm has a lot of defects when it isapplied to image segmentation, such as needing to confirm the number of cluster categories inadvance, quite sensitive to initial parameters, weak ability to resist noise, weak identificationability and easy to fall into local extremum to non-convex cluster. Aiming at the defects ofFCM algorithm, this paper studied intensively, and the main research summarized as follows.1) Aiming at the defect of week anti-noise capacity of FCM algorithm, this paperadopts the adaptive median filter and combines with the weighted FCM algorithm to conductiterative filtering segmentation, and then proposes the weighted FCM image segmentationalgorithm based on image filtering. Experiments show that the algorithm not only can restrainthe influence of noise but also has a good segmentation effect.2) This paper introduces the principal component transformation and analyzestwo-dimension histogram in the RGB color space to determine automatically the number ofcluster and initial value, and then proposes an adaptive color image segmentation algorithm.Experiments show that the algorithm not only makes up the deficiency of the FCM algorithmto image segmentation but only has a content segmenting effect to color image.3) This paper introduces the principal component transformation to multicenter imagesegmentation algorithm to determine automatically the number of cluster in the RGB colorspace, and spectral clustering and transitive closure to construct new similarity matrix, andthen proposes a color image segmentation method about FCM based on spectral clusteringand transitive closure. The effectiveness of the proposed method has been proved in the experiment. Meanwhile, compared with segmentation results gained by the traditional FCMclustering image segmentation algorithm, the results show that our method has a bettersegmentation effect.