Research For Image Segmentation Based On Fuzzy Clustering Analysis

Image segmentation is just to segment an image into different sub-images with different characters and get some interested objects. It is a key step from image process to image analysis, plays an important role in image engineering, and is applied in a lot of fields such as computer vision, image coding, pattern recognition, medical image processing and so on.Images themselves are very uncertain and inaccurate. It is found that fuzzy theory is able to give a good description of such uncertainties and image segmentation is just the classification of image pixels. In recent years, some experts are making efforts to apply the fuzzy clustering method in image segmentation, and it is more effective than the traditional image processing method. However, there are still some problems with classical image segmentation based on fuzzy clustering.Therefore, the paper first demonstrates the definition and the existing several algorithms of image segmentation, then on the basis of the fuzzy clustering theory, has drawn out HCM algorithm and FCM algorithm from the data set division, and a series of new ideas is presented in view of some drawbacks of FCM algorithm for image segmentation. The main research results can be concluded as follows:Firstly, 2D histogram is constructed by the denoising method based on adaptive median filtering. Then, with combining weighting FCM algorithm and pyramid structure, a new fast image segmentation method is proposed based on a 2D histogram weighting and pyramid structure FCM algorithm. The new algorithm overcomes some disadvantages of the standard FCM algorithm, for example lower speed and sensitivity to noise.Secondly, a cluster validity function, named modified partition fuzzy degree, is introduced for realization of automatically determining the optimal category number of image segmentation. Moreover, the effect of weighting exponent m in FCM algorithm on segmentation performance is investigated. Some meaningful conclusions are made which illustrate the relations between image features and parameter m, and connection between segmentation precision, speed and m.Finally, from the point of view of neighboring membership constraint, a new clustering objective function is proposed, which results in FCM algorithm for image segmentation based on neighboring membership constraint. Experiments of the proposed algorithm on synthetic test image and realistic image prove its different result along with different penalty coefficient.