Research On Image Segmentation Methods Based On Fuzzy Clustering

Image segmentation is the basic-level part of computer vision theory and hasvery important research content which feature extraction and target recognition arebased on. The image segmentation quality will directly affect the feature extractionand object recognition. Image segmentation aims to understand the content of theimage and extract the object of interest in accordance with the requirements of thespecific application, such as edges, regions, etc. Image segmentation method basedon clustering algorithm had been proposed earlier. As the introduction of fuzzy settheory, fuzzy clustering algorithm is a combination of both the advantages of theclassic theories, one of the most representative is fuzzy c-means clusteringalgorithm. As an unsupervised fuzzy clustering method, it does not require humanintervention in the implementation of the algorithm. Especially in dealing with thedigital image which existing the vagueness and uncertainty the method can reflectsthe advantages of fuzzy math. The method has been systematically studied andapplied to the areas such as image segmentation, object recognition, medicaldiagnosis,etc.This paper comprehensively describes the basic concepts of imagesegmentation and several typical image segmentation methods firstly, and focuses onthe theory of fuzzy clustering method which is combined of fuzzy theory andclustering method. Then, we made a detailed representation of the derivation processand execution steps of the fuzzy c-means clustering. Standard FCM-based imagesegmentation method still has inherently flaws such as computation complexity,initial cluster centers need to artificially specify which can affect the clustering resulton a certain extent, selection of smoothing factor m lacks theoretical guidance, verysensitive to noise and intensity inhomogeneities, etc. Due to the similarity distanceused in the clustering process, which only consider the Euclidean distance of pixelgray level characteristics, using the standard FCM image segmentation method to thenoisy image can not be satisfactory. A new distance had proposed to solve theproblem, which according to the mean filter principle in handling image noise andgiving full consideration to the pixel gray level information and neighborhood information, experimental results show that the method can segment imageeffectively with reducing the noise interference.To further enhance the noise immunity of the algorithm and the effect ofimage segmentation, the ability to effectively use spatial information is a key issue.This paper presents an adaptive fuzzy c-means clustering image segmentationmethod to take full advantage of the spatial position information of the pixels anddivide pixels into some different categories using the fuzzy inference system. Usingappropriate similarity distance for different types of pixel to cluster. Throughsynthetic images and natural images segmentation experiments it proved that theimproved method has stronger noise immunity and higher accuracy of segmentationcompared to traditional FCM method.