Research On The Image Segmentation Method Based On Fuzzy Clustering

As the basis of the object identification, motion analysis and scene reconstruction, image segmentation plays a role of link between the low level vision system and the high level vision system. Massive information is contained in images like surface's or texture features,which makes the processing of these kinds of images have been widely used in image analysis of remote sensing and medical field. It's of key importance how to describe image features with a proper model, while Fuzzy C-Means(FCM) clustering model possessed with complete theoretical basis and the ability of auto segmenting, which make it fit to describe the surface or texture image featured with uncertainty and fuzziness. Therefore, the image modeling and segmenting methods based on the FCM models were proposed in this paper, which focused on algorithm operating speed, anti-noise ability and the application of FCM on texture segmentation. The main lines can be classified into three sections:Firstly, a FCM image segmentation method based on the improved initiate parameter was raised. In which, a FCM cluster center initiate method was presented to solve the problems which brought by the random cluster center generated in original FCM method, and then a weighting coefficient of two-dimensional histogram was put forward by introducing the adjacent related space information of pixel on the basis of one-dimension histogram weighting method. We integrated this two-dimension histogram weighting coefficient with FCM model (2DWFCM), and the segmentation results was got through iterating process with the target function. Take the remote sensing image and medical image as examples, the experiment proved the effectiveness and the segmental accuracy of the method we proposed,compared with FCM and the one-dimension histogram weighting FCM methods.Then, to overcome the FCM clustering and the kernel FCM problems of being sensitive to noise and the long iteration time costs, the two-dimensional weighting coefficient was introduced into the Gaussian kernel FCM model and formed the 2DWGKFCM model. The inserting of Gaussian kernel makes the 2DWGKFCM model qualified not only the high segmentation efficiency, but also a better noise immunity and segmental accuracy behavior compared to the 2DWFCM. The experiment chose remote sensing images as targets, proofed the method we argued performed with a higher segmenting efficiency and a better accuracy in noise condition, compared to the origin kernel FCM or histogram weighting FCM models.At last, the FCM segmentation model based on the local binary pattern for texture image(LBPFCM) was presented for texture image segmentation and analysis. It's shown the local binary pattern (LBP) possessed with premium separating capacity, but its results is hard for naked eyes to recognize. As texture image is of uncertainty and fuzziness, we combined the FCM model with LBP, and chose the gray level co-occurrence matrix (GLCM) to analyze the segmentation results of the LBPFCM model. The final results proved that, applied in natural texture image and noise polluted irregular rock texture images, LBPFCM bear with a better image segmentation efficiency and accuracy in segmentation.